Deep learning-driven recognition of panoramic tumor microenvironment features in H&E sections and its application.

Deep learning-driven recognition of panoramic tumor microenvironment features in H&E sections and its application - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice J Immunother Cancer . 2026 Apr 9;14(4):e014429. doi: 10.1136/jitc-2025-014429 Search in PMC Search in PubMed View in NLM Catalog Add to search Deep learning-driven recognition of panoramic tumor microenvironment features in H&E sections and its application Han Zhang Han Zhang 1 Center of Clinical Big Data and Analytics of the Second Affiliated Hospital and School of Public Health, Zhejiang University School of Medicine, Hangzhou, Zhejiang, China 2 Zhejiang Key Laboratory of Intelligent Preventive Medicine, Hangzhou, Zhejiang, China Find articles by Han Zhang 1, 2 , Qinyi Huang Qinyi Huang 1 Center of Clinical Big Data and Analytics of the Second Affiliated Hospital and School of Public Health, Zhejiang University School of Medicine, Hangzhou, Zhejiang, China 2 Zhejiang Key Laboratory of Intelligent Preventive Medicine, Hangzhou, Zhejiang, China Find articles by Qinyi Huang 1, 2 , Bing Shang Bing Shang 1 Center of Clinical Big Data and Analytics of the Second Affiliated Hospital and School of Public Health, Zhejiang University School of Medicine, Hangzhou, Zhejiang, China 2 Zhejiang Key Laboratory of Intelligent Preventive Medicine, Hangzhou, Zhejiang, China Find articles by Bing Shang 1, 2 , Guowei Su Guowei Su 1 Center of Clinical Big Data and Analytics of the Second Affiliated Hospital and School of Public Health, Zhejiang University School of Medicine, Hangzhou, Zhejiang, China 2 Zhejiang Key Laboratory of Intelligent Preventive Medicine, Hangzhou, Zhejiang, China Find articles by Guowei Su 1, 2 , Huakang Tu Huakang Tu 1 Center of Clinical Big Data and Analytics of the Second Affiliated Hospital and School of Public Health, Zhejiang University School of Medicine, Hangzhou, Zhejiang, China 2 Zhejiang Key Laboratory of Intelligent Preventive Medicine, Hangzhou, Zhejiang, China Find articles by Huakang Tu 1, 2, ✉ Author information Article notes Copyright and License information 1 Center of Clinical Big Data and Analytics of the Second Affiliated Hospital and School of Public Health, Zhejiang University School of Medicine, Hangzhou, Zhejiang, China 2 Zhejiang Key Laboratory of Intelligent Preventive Medicine, Hangzhou, Zhejiang, China Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise. Additional supplemental material is published online only. To view, please visit the journal online ( https://doi.org/10.1136/jitc-2025-014429 ). No, there are no competing interests. ✉ Professor Huakang Tu; [email protected] Received 2025 Nov 26; Accepted 2026 Mar 26; Collection date 2026. Copyright © Author(s) (or their employer(s)) 2026. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ Group. This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See https://creativecommons.org/licenses/by-nc/4.0/ . PMC Copyright notice PMCID: PMC13084830  PMID: 41956540 Abstract The tumor microenvironment (TME), composed of tumor cells together with stromal cells, immune cells, vascular networks, and other components, constitutes a complex ecosystem that plays a decisive role in tumor initiation, progression, metastasis and therapeutic response. Traditional pathological diagnosis mainly relies on pathologists manually examining H&E-stained tissue sections under the microscope, a method that not only suffers from substantial interobserver variability but also has relatively low analytical efficiency. With the rapid development of computational pathology, the integration of whole-slide imaging technology and deep learning algorithms has provided powerful tools for characterizing tumor microenvironment. These techniques enable automated characterization of cellular, spatial and molecular heterogeneity within the tumor microenvironment, providing integrated insights that advance precision diagnostics and improve prediction of therapeutic response and patient outcomes. Based on a comprehensive review of existing research, this paper highlights recent advances in deep learning-driven recognition of panoramic TME features from H&E slides and their clinical applications and further discusses both the translational potential and current limitations of this technology in oncology research and clinical applications. Keywords: Tumor microenvironment - TME, Histology, Pathology Introduction The tumor microenvironment (TME) is a dynamic ecosystem of tumor, stromal and immune cells embedded within the extracellular matrix and vasculature. Interactions among these components create marked spatial and molecular heterogeneity that shapes tumor progression, immune escape, and clinical outcomes. 1 2 Deciphering and quantifying this heterogeneity is critical for understanding tumor biology and improving clinical strategies. Current approaches to evaluate TME heterogeneity can be broadly divided into morphology-based and sequencing-based methods. Traditional histopathology remains the foundation of TME assessment because it is low-cost and routinely available, but manual interpretation is time-consuming and prone to interobserver variability. 3 Clinically established molecular assays such as PCR and fluorescence in situ hybridization allow sensitive detection of selected markers but provide limited coverage of the complex cellular landscape. 4 High-throughput and single-cell sequencing approaches extend molecular profiling to genome-wide or transcriptome-wide scales, but their complexity, expense, and lack of spatial context limit routine clinical use. 5 6 Spatial transcriptomics has recently emerged to integrate molecular profiles with tissue architecture, although its clinical translation is currently limited by high cost, technical demands, and scalability. 7 In this context, computational pathology aims to bridge these approaches by integrating artificial intelligence with digital histology, thereby supporting objective and scalable analysis of TME composition and spatial organization. Computational pathology has developed rapidly with advances in whole-slide imaging (WSI) and artificial intelligence (AI), becoming an important tool for tumor diagnosis, grading, subtyping, and treatment response. 8 , 12 As research moved beyond diagnostic applications, early studies began to explore different ways of histopathology image analysis. Qualitative analysis refers to identifying what is present, recognizing specific cell types and tissue structures on H&E slides to outline the basic composition of the specimen. 8 , 1012 Quantitative analysis focuses on measuring how much is present, using these identified elements to calculate cell densities, tissue proportions or morphologic features such as immune-cell infiltration or extracellular-matrix content. 13 , 15 Localization analysis looks at where these components are situated, mapping their spatial distribution and relationships, for instance, the positioning of immune cells within lymph nodes or the nuclear organization that separates aggressive from indolent disease. 14 16 Integrative analysis brings these findings together by combining histologic, spatial and molecular information to relate tissue features to underlying biology or clinical outcome. 17 , 20 Although these approaches were not initially developed to study the TME, they laid the foundation for applying computational pathology to study tissue architecture, cellular composition and spatial organization. Building on this foundation, this review summarizes recent progress in computational pathology for TME analysis, emphasizing qualitative, quantitative, localization and integrative analyses and their clinical applications ( figure 1 ). Figure 1. Overview of deep learning-driven panoramic tumor microenvironment analysis in H&E slides and clinical applications. This figure provides an overview of the computational pathology framework for TME analysis based on H&E slides. The framework encompasses qualitative, quantitative, localization, and integrative analyses. These analyses contribute to clinical applications, including pathology-assisted diagnosis, prognostic assessment, immune status evaluation, and treatment decision guidance. CNN, convolutional neural network; GNN, graph neural network; TME, tumor microenvironment; ViT, Vision Transformer. Open in a new tab AI-driven qualitative analysis of the tumor microenvironment The composition and organization of cellular and stromal elements within the TME are critical for understanding tumor heterogeneity, assessing immune status and guiding therapeutic decisions. Conventional histopathology provides essential morphological information from H&E slides, yet the manual evaluation of complex cellular patterns remains inefficient and variable. In recent years, the rapid development of computational pathology has provided promising tools to address this issue. Among these approaches, qualitative analysis represents the foundational step, focusing on the identification and characterization of distinct cell types and tissue structures to delineate the basic histologic composition of the specimen. Early computational pathology studies used digital morphology to examine how histologic patterns relate to clinical information. Beck et al reported stromal patterns in breast cancer associated with outcome, and Basavanhally et al characterized lymphocytic infiltration in HER2-positive tumors using nuclear morphology. 13 20 These studies showed that routine H&E slides contain informative visual signals but did not yet perform explicit cell-type labeling. A methodological shift toward direct cellular identification began with CRImage in 2012, which automatically classified tumor, lymphocyte and stromal cells in The Cancer Genome Atlas (TCGA) breast cancer cohorts, establishing an initial three-class paradigm for cell recognition on H&E. 21 The development of curated datasets and recognition models soon made cell typing more consistent across studies. PanNuke annotated five cell categories (neoplastic epithelial, non-neoplastic epithelial, inflammatory, connective and dead cells) across multiple organs; MoNuSAC provided nuclear-level labels for epithelial cells, lymphocytes, macrophages and neutrophils; and CoNIC defined six cell categories (epithelial, lymphocyte, neutrophil, plasma, eosinophil and connective cells) combining segmentation and classification tasks. 22 , 25 Based on these resources, researchers developed models such as Stardist, HoverNet and Cellpose to achieve precise nuclear segmentation and cell recognition, while transformer-based frameworks further improved adaptability across tissue types and staining conditions. 26 , 30 Building on the early three-class paradigm, cell recognition has expanded to include more diverse populations within the TME, such as stromal cells (including fibroblasts, adipocytes, and endothelial cells) and immune cells (including eosinophils, neutrophils, macrophages, and plasma cells). 31 Beyond cell-type identification, already recognized cell populations can be further subdivided into subtypes based on their interactions or relationships within the tissue. For example, adipocytes can be classified as cancer-associated or distant adipocytes according to their distance from cancer cells or fibrotic regions in breast cancer. 32 The registration of H&E slides with multiplex immunofluorescence (mIF) has further enabled the construction of high-resolution datasets that integrate morphological and immune-marker information, generating over 1.1 million accurately annotated cells for refined single-cell reference labeling. 33 At the regional level, qualitative analysis has extended from cell-type identification to the delineation of major histologic compartments. Public datasets such as National Center for Tumor Diseases colorectal cancer dataset (NCT-CRC) and Harbin Medical University gastric cancer dataset (HMU-GC) provide region-level annotations including tumor epithelium, normal epithelium, stroma, inflammatory, muscle, mucosa, necrosis and adipose regions, establishing a basis for standardized segmentation. 34 35 In parallel, deep convolutional neural network-based models have achieved reproducible epithelial-stromal delineation in breast cancer, while another study in lung cancer further refined tissue characterization by distinguishing not only tumor and stromal areas but also specialized pulmonary regions such as bronchial epithelium, vasculature, glands, and peribronchial cartilage. 36 37 Meanwhile, MSegNet advanced qualitative analysis by identifying six peritumoral compartments in hepatocellular carcinoma, such as fibrotic reaction and microvascular invasion. 38 Collectively, these studies show that qualitative recognition now captures both cellular composition and tissue organization, offering a more comprehensive description of TME architecture. Integration of multiple tissue elements within single qualitative frameworks represents the next stage of development. The identification of tumor-infiltrating lymphocytes (TILs), which involves concurrent recognition of tumor and lymphocyte cells, has become a key approach for evaluating immune involvement within the TME. Extensions of CRImage and commercial tools such as HALO have incorporated automated TIL detection to jointly recognize epithelial, stromal and immune cells. 39 40 Building on these efforts, TILScout generated automated TIL maps across 28 cancer types, linking qualitative TIL recognition with survival endpoints and demonstrating its potential for prognostic stratification. 40 CollaTIL extended this framework to jointly recognize immune cells and collagen fibers, suggesting that stromal structure contributes to immune regulation. 39 In high-grade serous ovarian cancer, computational pathology was used to qualitatively compare tissues collected before and after neoadjuvant chemotherapy (NACT), linking TIL and collagen features with treatment response. 41 Robust qualitative identification of cells and tissue regions forms the basis for subsequent quantitative, localization and integrative analyses of the TME. On this basis, studies such as smsTIL characterized TIL morphology in triple-negative breast cancer (TNBC), demonstrating its prognostic relevance. 42 Automated cell recognition has also supported descriptive classification of immune patterns across tumor types, for example, distinguishing immune-rich from immune-poor contexts directly on H&E slides. 43 , 45 Frameworks including CRImage, HALO, CollaTIL and TILScout further standardized cell-level and region-level characterization, and were later adopted for more detailed modeling of tumor-stroma relationships and immune organization. 40 46 47 In summary, qualitative analysis has progressed from descriptive morphology to automated, data-driven identification of cellular and regional components supported by large datasets and modern algorithms ( figure 2 ; online supplemental table 1 ). By enabling the simultaneous recognition of tumor, immune and stromal features within routine H&E slides, these frameworks bridge traditional histopathology and computational modeling, and now form the conceptual foundation for future systematic analysis of the TME. Figure 2. Qualitative analysis of the tumor microenvironment in H&E slides. This figure illustrates qualitative analysis of the tumor microenvironment, which involves cell-level and tissue-region classification and segmentation based on H&E-stained slides. Several publicly available annotated datasets, such as PanNuke, MoNuSAC, CoNIC, NCT-CRC, and HMU-GC, have been established to support the development and validation of these analyses. *Representative cellular and regional recognition examples are adapted from previously published studies. 177 178 . Open in a new tab AI-based quantitative analysis of the tumor microenvironment Building on qualitative characterization of tissue composition, quantitative analysis aims to translate histologic observations into measurable parameters by systematically quantifying cellular, structural and morphometric features. A variety of tools, such as Mask R-CNN, ConvPath and QuPath, have been developed for automated recognition of tumor, stromal and immune cells. 48 , 50 These user-friendly tools enable the construction of multicell-type maps across pan-cancers. 49 As early as 2014, researchers applied machine learning to quantify vascular patterns in breast cancer H&E slides and showed their correlation with estrogen receptor expression. 51 With growing insight into the TME, quantitative indicators have become increasingly diverse and can be broadly classified into two categories: proportion-based indicators, which capture ratios between different cell types or tissue regions (eg, the Tumour-Lymphocyte Spatial Interaction score (TLSI-score) reflecting tumor-lymphocyte interactions in breast cancer); and count/density-based indicators, which include both the absolute number of specific cells or structures and their normalized density relative to regional area. 52 Powered by deep learning, the quantification of these indicators has been automated and incorporated into a systematic framework for TME analysis. Their clinical relevance has been demonstrated across multiple cancer types, including the association of the relative composition of tumor, stroma and lymphocytic regions and mitotic density with pathological complete response (pCR) after neoadjuvant chemotherapy in breast cancer, the association of stromal composition and tumor cell density with patient survival risk in malignant mesothelioma, and the association of periductal lymphocyte density with recurrence risk in ductal carcinoma in situ. 53 , 55 Representative proportion-based and count/density-based indicators are discussed in the following sections. The stromal compartment of the TME, consisting of connective tissue, vasculature and extracellular matrix, plays a critical role in tumor growth and progression. The tumor-stroma ratio (TSR) is a key determinant of tumor aggressiveness, immune evasion and therapeutic response. 56 Early TSR assessment depended on manual inspection of H&E slides to estimate epithelium and stroma proportions, a process limited by low efficiency and poor reproducibility. With the advancement of digital pathology and AI, deep learning-based automated methods have gradually replaced manual approaches. In 2020, regional annotations from the NCT-CRC-HE-100K dataset were used to train a VGG-19 model, representing the first automated approach for TSR quantification in TCGA colorectal cancer WSIs. 35 57 When combined with clinical features, this model achieved accurate prognostic prediction. 57 Subsequent studies extended automated TSR quantification to breast and early rectal cancers and incorporated additional TME features, such as necrosis and lymphocyte infiltration, providing a more robust framework for exploring TME heterogeneity. 58 , 60 Based on these advances, in 2022, Jakab et al developed a user-friendly commercial software for TSR calculation to assist pathologists in colorectal cancer risk stratification. 61 Tumor-infiltrating immune cells, including TILs such as T cells and B cells, as well as tumor-associated macrophages, can exhibit context-dependent roles within the TME, contributing to immune surveillance while also promoting tumor progression under certain conditions. 62 63 Automated TIL quantification was first developed using a convolutional neural network (CNN) model trained on CD45-IHC-annotated breast cancer slides. This model distinguished immune-rich from immune-poor regions with performance comparable to expert pathologists and laid the foundation for large-scale spatial analyses. 45 56 Subsequent TCGA studies expanded this approach across multiple cancer types and generated pan-cancer TIL spatial maps integrating histology, gene expression and immune subtypes, marking a milestone in digital TME quantification. 64 Furthermore, commercial platforms such as CRI iAtlas and Lunit SCOPE IO enabled online TIL analysis and automated classification of spatial immune phenotypes (inflamed, excluded and desert). 65 , 67 Lunit SCOPE IO has shown predictive value for immune checkpoint inhibitor (ICI) response in non-small cell lung cancer (NSCLC) and provides reproducible stromal-TIL scoring in breast cancer, which has been applied to assess response to neoadjuvant therapy. 65 68 The platform has now been adopted for more than 20 cancer types and supports the use of TIL-based metrics as biomarkers for ICI response. 69 , 71 In contrast to TILs, quantitative analysis of tumor-associated macrophages from H&E images remains more challenging. This is partly due to greater morphological heterogeneity and fewer nucleus-level annotations of macrophages in commonly used training datasets, resulting in less consistent classification performance and poorer whole-cell (instance) segmentation quality compared with lymphocytes. 25 72 Recent work has begun to address this limitation by leveraging virtual immunohistochemistry (IHC) approaches; for example, Aggarwal et al developed a framework that infers CD163+ from H&E slides and enables identification of M2-TAM density in human papillomavirus-associated oropharyngeal squamous cell carcinoma (HPV+ OpSCC). 73 Tertiary lymphoid structures (TLS), organized immune aggregates within the TME, are closely associated with antitumor immunity and prognosis. Automated TLS identification generally follows two strategies. One strategy infers TLS by detecting clusters of TILs and refining their boundaries, whereas the other directly classifies image patches as TLS or non-TLS based on supervised learning from expert annotations. 74 , 77 These strategies have enabled quantitative TLS assessment and facilitated prediction of prognosis and immunotherapy response. 74 78 Extending these approaches, recent work in lung cancer incorporated tissue area measurements to derive the TLS density across the whole tumor and the TLS-to-necrosis area ratio as prognostic indicators. 37 Other TME features have further broadened the scope of quantitative research. Recent studies in colorectal and lung cancers developed progression-free survival prediction models that incorporated ratios such as tumor fibrosis ratio, tumor inflammation ratio, tumor vessel ratio, tumor necrosis ratio and tumor background ratio, indicating the expanding scope and future promise of computational pathology in quantitative TME analysis. 79 80 In summary, computational pathology has made substantial progress in quantitative TME analysis such as cell mapping, TSR measurement and TIL/TLS evaluation ( figure 3 ; online supplemental table 2 ). Deep learning has further improved the accuracy and efficiency of these analyses, advancing the evaluation of immune response, tumor aggressiveness and therapeutic outcomes. Figure 3. Quantitative analysis of the tumor microenvironment in H&E slides. This figure illustrates a quantitative analysis of the TME using two types of indicators: proportion-based indicators such as TSR and other TME ratios, and count/density-based indicators represented by TIL and TLS quantification. *Immune phenotyping and TLS quantification examples are adapted from prior studies. 74 77 93 CE, cancer epithelium; CS, cancer stroma; TIL, tumor-infiltrating lymphocyte; TLS, tertiary lymphoid structures; TME, tumor microenvironment. Open in a new tab AI-driven localization analysis of the tumor microenvironment The spatial placement of cells within the TME shapes their participation in biological processes and substantially influences histopathological interpretation and functional assessment. A growing body of evidence indicates that cellular location and spatial distribution within the TME often conveys more information than overall abundance. For example, CD8+ T cells in the distant stroma independently predict disease-specific survival in breast cancer, whereas dense infiltration of CD3+ T cells at the invasive margin is closely associated with disease-free survival (DFS) in colorectal cancer. 81 82 Localization analysis focuses on characterizing the spatial arrangement and interactions of distinct cellular components, mapping their distribution patterns within the tissue to reveal microenvironmental organization and functional relationships. Computational pathology is transforming histological assessment by integrating deep learning with image analysis, enabling more precise characterization of cellular localization and cell–cell interactions within the TME, for example, Ceograph associates cell spatial features, such as distribution, morphology, proximity, and interactions, with malignant transformation risk in oral potentially malignant disorders and treatment response in lung cancer. 83 Recent studies increasingly combine automated segmentation of tissues such as tumor and stroma with spatial mapping of lymphocytes to support prognostic assessment. In head and neck squamous cell carcinoma (HNSCC) and intrahepatic cholangiocarcinoma (ICC), accumulating evidence indicates that patient survival is associated with the spatial distribution of lymphocytes rather than lymphocyte density alone. Specifically, survival benefit in HNSCC has been linked to lymphocyte spatial co-occurrence with tumor-associated stroma quantified by the tumour-associated stroma infiltrating lymphocytes (TASIL) score, while in ICC, survival is associated with graph-derived lymphocyte features, including cluster heterogeneity, local neighborhood irregularity and peritumoral lymphocyte density. 84 85 Based on this concept, an AI-driven H&E image analysis platform was developed for colorectal cancer (CRC), enabling automated tissue segmentation, cell-type classification and extraction of spatial distribution features. By integrating these image-derived features with transcriptomic and clinical data, the platform facilitates microsatellite instability (MSI) assessment and survival prediction. 86 This integrative strategy streamlines large-scale analysis and provides a new computational framework for precision oncology. Similar frameworks have been extended to other cancer types. For instance, deep learning models such as ARA-CNN (Accurate, Reliable and Active convolutional neural network) segment lung cancer H&E slides into multiple tissue compartments and quantify the intratumor lymphocyte ratio, a metric capturing the spatial relationship between tumors and immune aggregates and revealing TME heterogeneity. 87 Similarly, in patients with NSCLC, another study showed that spatial features describing cell–cell connectivity and relative positioning among stromal cells, fragmented nuclei and red blood cells were significantly associated with overall survival. 88 Furthermore, another study developed HistoTIL for ICI response prediction in patients with NSCLC and gynecological cancers by extending conventional TIL density assessment. In addition to nuclei morphometrics and TIL abundance, HistoTIL incorporated spatial descriptors capturing the cellular composition and nuclear morphology within TIL-centered neighborhoods, tumor-TIL proximity and relative spatial positioning, and cluster-level tumor-immune relationships, including the area and density ratios between tumor clusters and surrounding TIL aggregates. 89 Biomarker-directed IHC and immunofluorescence (IF) enable in situ, single-cell-level phenotypic and functional characterization of the TME at the protein level while preserving tissue architecture. 90 Combining H&E-based localization with IHC or IF enables more detailed identification of cellular subtype, particularly for immune cell phenotyping. This strategy bridges morphological and molecular information, providing complementary insights into the tumor immune landscape. Early studies applied deep learning for tumor segmentation and, together with IHC, quantified CD8+ T cells distribution, which predicted response to programmed cell death protein 1 therapy. 91 Subsequent studies incorporated manual annotation or nuclear morphology-based region segmentation with automated IHC cell counting to map the spatial distribution of T-cell subsets and to classify immune phenotypes, such as hot, cold and excluded, revealing prognostic relevance in ovarian and CRCs. 92 93 Collectively, these studies have established a systematic workflow that combines tissue segmentation, IHC-based cell labeling and spatial statistical modeling to quantify immune cell density, clustering and heterogeneity. Using this pipeline, immune cell density at the invasive margins has been implicated as a marker of immune response in TNBC. 94 Recently, commercial platforms like HALO have standardized this workflow, supporting integration of programmed death-ligand 1 scoring, TIL density and tumor mutational burden (TMB) in clinical applications. 95 , 97 To further reduce the reliance on additional staining in routine workflows, emerging computational pathology studies are exploring the direct inference of IHC, mIF, and other special stains from standard H&E WSI. 98 , 100 These virtual staining approaches draw on recent advances in generative adversarial networks, contrastive learning and diffusion-based models to estimate spatial biomarker expression (such as HER2, CD3, CD8 and CD163) from morphological features alone, potentially reducing tissue consumption and experimental burden. 73101 , 103 Although these methods remain largely at the proof-of-concept stage, ongoing efforts in large-scale, multicenter validation and the development of pathology foundation models are gradually moving H&E-based virtual staining toward future translational applications in TME analysis. 104 , 106 Computational pathology now provides essential tools for localization analysis within the TME ( figure 4 ; online supplemental table 3 ). Current work extends into multi-omics, integrating proteomics, DNA methylation and single-cell transcriptomics to examine the relationships among immune phenotypes, functional states and molecular traits. In parallel, advances in automated IHC analysis systems and H&E–IHC/mIF registration techniques enable more precise spatial annotation of immune markers, providing high-quality labels that support localization analysis of TME on H&E slides and promote the development of H&E-based virtual IHC/mIF models. 101 105 107 108 Together, these advances establish a robust H&E-based framework for TME research and broaden the role of pathology images in immune feature exploration. 109 110 Figure 4. Localization analysis of the tumor microenvironment in H&E slides. This figure illustrates the localization analysis of the TME, which is performed following cellular and regional recognition to describe cell distribution and spatial associations. Cell distribution analysis shows how different cell types are distributed across histologic regions (eg, tumor and stroma). Neighborhood-based spatial analysis summarizes the composition of neighboring cells around a given cell type, reflecting immune infiltration patterns. Cell graph-based spatial analysis further characterizes spatial relationships among cells using nucleus-to-nucleus distances to capture higher-level tissue architecture. Together, these localization analyses enable interpretation of TME heterogeneity through characteristic patterns of cell-region organization, cell–cell distances, and region-specific cellular distributions, which collectively reflect immune microenvironment states in H&E images. *Representative pathology images and spatial analysis schematics are adapted from previously published studies. 72 83 85 112 TASIL, tumour-associated stroma infiltrating lymphocytes; TME, tumor microenvironment. Open in a new tab AI-driven integrative analysis of panoramic tumor microenvironment features While qualitative, quantitative and localization analyses capture different aspects of the TME such as cell composition, abundance and distribution, these information is traditionally been processed in isolation. Integrative analysis has emerged as an important approach to uncover the complex biological mechanisms within the TME by combining histopathologic features with multi-omics and multimodal data to establish comprehensive correlations between tissue morphology, molecular alterations and clinical outcomes. By integrating information on cell phenotypes, density and localization, immune features can be characterized across both local and whole-tissue scales. Furthermore, combining histopathological features with molecular data such as transcriptomics and mutation burden enables investigation of the links between tissue morphology and molecular alterations. Moreover, multimodal integration strategies that incorporate imaging, omics and clinical data within unified analytical frameworks promise to improve the robustness and clinical applicability of TME characterization. Integrative analysis of qualitative, quantitative, and spatial features In recent years, researchers in computational pathology have developed a series of frameworks that integrate cell-level information, tissue architecture and spatial interactions, enabling multidimensional profiling of TME functional states. These frameworks not only enhance the biological interpretability of image analysis but also facilitate the discovery of clinical biomarkers and the development of personalized therapies. On one hand, studies increasingly incorporate refined spatial and structural image features. For example, a recent study proposed an analytical framework focused on “Human-Interpretable Image Features (HIFs).” Using the TCGA database, the authors systematically extracted 607 cell-level and tissue-level variables to build prediction models for immune checkpoint expression and clinical outcomes, covering features ranging from basic cell counts and tissue areas to more complex spatial indicators such as tissue interfaces, morphological patterns and cell proximity relations. 72 Another study focusing on HPV+OpSCC developed an open-source high content image analysis (HCIA) tool to extract multilevel features including the spatial distribution and nuclear morphology of tumor and immune cells from H&E images, helping identify patient subgroups suitable for treatment de-escalation. 111 On the other hand, research focus has transitioned from expert annotation-driven features identification to model-driven extraction of local image features through autonomous learning. Researchers applied a self-supervised learning model to automatically extract “Histomorphological Phenotype Clusters (HPC)” from unannotated H&E slides in the TCGA colon adenocarcinoma cohort (TCGA-COAD) to construct prognostic classifiers. This approach achieved excellent predictive performance in the AVANT large clinical trial dataset, demonstrating the potential clinical utility of integrated pathology image models in supporting treatment decision-making. 112 Integrating qualitative, quantitative and localization analyses broadens the scope of TME research by bridging traditional pathological feature extraction with high-dimensional analysis driven by clinical applications, establishing a foundation for multi-omics and multimodal integration. Integrative analysis of TME from morphological features and multi-omics Integrating multi-omics data, including genomics, transcriptomics and proteomics, with pathological images enables comprehensive characterization of the TME from both morphological and molecular perspectives. Public databases, such as TCGA and The National Cancer Institute's Clinical Proteomic Tumor Analysis Consortium (CPTAC), link H&E slides with matched multi-omics and clinical information, allowing integrated analyses of TME-associated molecular subtypes, mutational profiles, immune and stromal components, and their clinical relevance. 113 114 These resources have substantially advanced understanding of TME biology and its impact on tumor classification and patient outcomes. Supervised models In early work, researchers integrated H&E slides with TCGA transcriptomic and clinical data, focusing on nuclear morphology to examine associations with prognosis, molecular subtypes and biological processes. 17115 , 119 These studies primarily targeted tumor regions without resolving cell types and paid limited attention to spatial features of the TME. With advances in deep learning, studies have increasingly used tile-based representations and expert annotations to construct spatial graphs and have trained graph neural network (GNN) to predict multi-omics features such as KRAS/TP53 mutations, ZFHX4 expression and DNA methylation, demonstrating robust pan-cancer performance. 120 121 These strategies are relatively interpretable and highlight the contribution of TME components such as TILs and fibroblasts in predicting treatment responses and mutational status; however, their heavy reliance on expert annotations limits clinical applicability. 122 123 Weakly supervised and unsupervised models To reduce annotation burden, weakly supervised and unsupervised approaches have been increasingly adopted, using slide-level labels such as mutation status, methylation levels and clinical outcomes. MSI prediction is a flagship application of weakly supervised learning in computational pathology. Since the first MSI classifier for gastric cancer, subsequent studies have introduced architectures such as ShuffleNet, ResNet50 and GNN, improving generalization across multicenter, multi-ethnic cohorts and achieving AUCs (areas under the receiver operating characteristic curves) over 0.89, approaching clinical-grade performance. 124 , 126 Automated platforms such as MSIntuit and STAMP further support clinical translation by identifying MSI status directly from H&E slides without additional staining or sequencing. These pipelines have been reported to improve screening efficiency while maintaining high sensitivity, thereby supporting treatment stratification in gastrointestinal cancers. 127 128 Beyond MSI, weakly supervised and unsupervised models identify TME structural features and infer diverse omics states from H&E slides. Using weak labels including immune subtypes, Tumor Immune Dysfunction and Exclusion (TIDE) scores and CIBERSORT-inferred cell fractions, and architectures such as ResNet, Transformer and contrastive learning, these models predict TMB, HPV status, hypoxia phenotypes, immune subtypes and precancerous lesions. 129 , 137 In cohorts such as TCGA bladder urothelial carcinoma (BLCA) and CRC, these models achieve high accuracy without pixel-level annotations. When paired with multi-omics labels, they can detect TME morphological features overlooked by routine pathology, for example, epithelial and macrophage alterations associated with hypoxic breast cancer. 137 Foundation models in computational pathology Building on weakly supervised and unsupervised advances, current efforts concentrate on two directions. First, more efficient architectures are being developed to enhance image feature learning. For example, ETMIL-SSLViT, which combines Vision Transformer with self-supervised learning, has accurately predicted TMB status and pathological subtypes in endometrial and CRCs. 138 Contrastive-clustering models like CAMIL have predicted continuous transcriptomic, methylation and immune-related biomarkers across pan-cancers. 139 Second, general-purpose pathology foundation models like CTransPath, UNI, Virchow2 and GigaPath enable extraction of transferable TME features from H&E slides that support diverse downstream tasks, including image retrieval, mitosis recognition and WSI classification. 140 , 143 Leveraging such foundation models, prediction of key glioma biomarkers (IDH, TERT and 1p/19q) has reached mean AUCs around 0.94. 144 Additionally, STAMP, built on CTransPath, provides a weakly supervised platform that integrates the entire end-to-end modeling pipeline, from problem definition to evaluation, for solid-tumor biomarker prediction. Deployed across multiple centers, it supports molecular-state recognition across cancers (eg, MSI), illustrating the scalability and standardization potential of foundation model-based workflows in real-world settings. 128 Spatial transcriptomics–pathology mapping Spatial transcriptomics (ST) preserves spatial context that is lacking in conventional high-throughput sequencing approaches and extends transcriptomic profiling into the spatial dimension, enabling in situ analysis of gene expression patterns, cell state heterogeneity, and inferred cellular interactions within the TME. 145 Recent research has explored the correspondence between histological features and transcriptional states to infer spatial gene expression and cellular organization from H&E slides. 146 Frameworks such as stLearn, BayesSpace, SpaGCN, STAGATE and GIST use spatial positioning, graph structures or attention mechanisms, and in some cases incorporate histological feature extractors pretrained on large-scale histopathology cohorts to refine the spatial resolution of gene expression and improve the delineation of biologically relevant tissue regions. 147 , 152 Among these, STimage facilitates inference of spatial gene markers and cell-type distributions from routine H&E images, providing an interpretable link between tissue morphology and underlying cellular organization. 147 More recently, the OmiCLIP, trained on Visium data, maps ST onto pathology images to provide cell-level transcriptomic annotations, enabling more accurate interpretation of TME features from H&E images. 153 Although H&E-based prediction of spatial gene expression shows substantial potential for TME characterization, current methods remain in their early stage of development. Morphologic features are not solely determined by transcriptional variation, and their relationship with spatial gene expression is often partial or context-dependent. In combination with the high dimensionality and inherent noise of both imaging and transcriptomic data, as well as uncertainty in target gene selection, these factors limit robustness and clinical translation. 154 In conclusion, advances in H&E-based computational pathology now allow comprehensive characterization of TME heterogeneity, immune status and functional phenotypes through the integration of weakly supervised learning, pathology foundation models and multi-omics data, providing a framework that links tissue morphology with molecular mechanisms. Multimodal integrative analysis Multimodal integrative models have shown promising performance in predicting tumor treatment responses. Early frameworks such as NEXT combined H&E image texture features with RNA sequencing data to estimate immune cell proportions, generating predictions for B cells and other subpopulations that showed high concordance with IHC staining results. 155 Building on these advances, more sophisticated multimodal frameworks such as Pathomic Fusion and ENLIGHT-DeepPT integrate WSIs with transcriptomic, mutation and clinical data to predict therapy responses across treatment modalities and have demonstrated robust and generalizable performance across multiple cancer types. 156 157 Furthermore, a study of papillary thyroid carcinoma (PTC) incorporated pathological images, genomic profiles and immune scores to stratify risk and identify high-risk tumor regions associated with lymph node metastasis and DFS. 158 Using the largest breast cancer NACT dataset to date, DLMM (deep learning-based multi-task model) and MIFAPS (multimodal integrated fully automated pipeline system) integrated WSIs, clinical variables and MRI to predict NACT response and pCR, outperforming models based on pathology or imaging alone. 159 160 Similarly, in another independent NACT cohort, an integrated TME score combining H&E-derived inflammatory features with DNA, messenger RNA and clinical variables achieved high predictive accuracy for NACT response and showed robust generalizability in external validation. 161 These findings indicate that H&E-centered multimodal pipelines hold strong potential for clinical applications. AI-driven integrative analysis of the TME is progressing toward greater predictive performance, interpretability and transferability across tasks. Current efforts focus on developing predictors that generalize across tumor types, feature scales, modalities and omics, reshaping model design and informing the development of AI-assisted clinical decision support ( figure 5 ; online supplemental table 4 ). Figure 5. Integrative analysis of the tumor microenvironment in H&E slides. This figure illustrates integrative analysis of the TME, including multidimensional integration of qualitative, quantitative and localization analyses; multi-omics integration that associates histological features with genomic, transcriptomic, metabolomic and spatial transcriptomic profiles to infer molecular characteristics; and multimodal integration that combines histopathology with clinical, radiological and molecular data. These integrative analyses enable more comprehensive and detailed characterization of the TME. HPV, human papillomavirus; MSI, microsatellite instability; TMB, tumor mutational burden; TME, tumor microenvironment. Open in a new tab Conclusion and perspectives Accumulating evidence indicates that computational pathology not only captures tissue architecture but also provides a practical and cost-effective alternative for molecular subtyping, prognostic stratification and therapy-response prediction. This review summarizes recent advances in panoramic TME characterization from H&E slides across four domains: qualitative, quantitative, localization and integrative analysis, highlighting representative studies on immune heterogeneity, functional-state assessment, spatial organization and image–omics integration. The broader application of deep learning, particularly pathology foundation models, is shifting the role of H&E slides from a supplementary tool to a central platform for investigating tumor biology and guiding precision treatment. Despite significant technical progress, challenges remain in TME interpretation. In qualitative analysis, distinguishing immune cell subtypes remains difficult, as widely used public datasets often contain only broad categories (eg, tumor cells, lymphocytes), while immune responses depend on specific functional subsets (eg, CD4+T, CD8+T and regulatory T cells). Functional states such as activation or exhaustion cannot be reliably determined from H&E slides alone. Annotations at this resolution typically require IHC or single-cell/ST, but cost and limited accessibility restrict their application in large cohorts. 162 Even well-established TIL pipelines can misclassify non-specific inflammation as tumor-associated infiltration in challenging contexts (eg, secondary follicles), leading to false positive results. 74 Developing methods that maintain identification accuracy while capturing functional information under limited annotation remains a key priority for future work. Quantitative analysis provides a standardized framework for describing TME features, but several limitations remain. Most studies have concentrated on indicators such as TIL density and TSR, whereas other histopathological indicators, including microvessel density, fibrotic and necrotic area fractions, and the proportions of adipose tissue and collagen-rich or desmoplastic stroma, have received less attention. In addition, measurement procedures differ among centers and tissue types, and the absence of unified normalization and reporting standards limits comparability across cohorts. 163 Quantitative indicators based solely on cell counts or regional proportions also overlook spatial organization, as similar overall values can reflect distinct spatial patterns related to different immune states or biological behaviors. 164 Future studies should integrate quantitative indicators with spatial descriptors and evaluate their robustness in multicenter datasets to improve reproducibility and consistency across studies. Localization analysis provides a framework for quantifying and interpreting the spatial organization of cellular populations and tissue structures within the TME. However, important interfaces, such as the tumor-stroma boundary, tumor-vascular contacts and epithelial-immune fronts, have received comparatively less attention, and interface-derived metrics remain sensitive to boundary ambiguity, intercenter annotation variability, variations in staining and imaging protocols, and scale dependence. 165 Beyond interfaces, localization analysis can delineate necrotic areas, regions of dense immune infiltration, stromal subtypes and vascular architecture. Quantifying these spatial structures provides valuable insights into lymphocyte trafficking, angiogenesis and potential metastatic routes, yet these analyses continue to face challenges related to segmentation accuracy, threshold selection and consistency across scales. Future studies may improve localization precision by integrating patch-level and whole-slide analyses across magnifications. In addition, incorporating advanced computational approaches, such as GNNs and spatial attention mechanisms, could enhance the modeling of spatial dependencies and reduce prediction errors. Integrative analysis expands the scope of TME assessment but still faces several limitations. Many studies rely on slide-level labels (eg, mutations, hypoxia) to guide regional feature extraction, but such labels lack cell-level detail and clear biological context, thereby weakening associations with immune phenotypes, molecular subtypes and clinical outcomes. In practice, H&E slides and multi-omics data are often collected at different times or from different tissue blocks, creating batch effects and tissue-state discrepancies that complicate data alignment. High-dimensional omics combined with small cohorts increases the risk of overfitting and spurious correlations. 166 167 Furthermore, real-world cohorts frequently contain incomplete data, with some patients lacking omics while retaining imaging or clinical information, placing higher demands on the robustness of multimodal approaches. To address these challenges, future efforts should focus on building multicenter, paired H&E–multi-omics datasets with regional labels derived from spatial omics. Developing pathology foundation models jointly trained on morphology and multi-omics may further enhance generalization and biological interpretability across cancer types and clinical settings. In addition, integrative methods should be robust to missing modalities to maintain performance and clinical applicability in real-world settings. 168 However, moving from methodological development to clinical application faces additional challenges. Model generalizability across centers remains restricted, as differences in slide preparation, staining protocols, and scanner platforms lead to inconsistent performance in external cohorts and reduced diagnostic reproducibility. Quality control and normalization pipelines such as GrandQC and HistoQC reduce technical artifacts and staining variability, but do not fully resolve intercenter differences. 169 170 In routine clinical practice, these challenges are further shaped by imbalanced data distributions, including heterogeneous TME composition, substantial interpatient variability, and rare histological patterns that are under-represented in research datasets, which limits robustness. 171 At the model level, the black-box nature of deep learning further restricts interpretability and reduces pathologists’ confidence in algorithmic outputs. 171 More fundamentally, many approaches infer immune states, molecular expression, or functional phenotypes from H&E morphology based on statistical associations rather than mechanistic causality, limiting their use as independent surrogates for molecular assays and safe extrapolation across tumor types, treatment contexts, and disease stages. 172 In addition, the large data volume and computational demands of WSI analysis complicate integration into existing hospital information systems. 173 Finally, the clinical role of computational pathology remains unclear, as it is often unspecified whether AI systems are intended to replace, triage, or quantitatively augment pathologist assessment, leading to ambiguity in workflow integration and responsibility allocation. Nevertheless, advances in AI provide new opportunities for TME exploration. Serial-section three-dimensional reconstruction can recover vertical architecture and capture small-scale features such as tumor budding, whereas virtual staining reduces laboratory costs and supports rapid diagnostic workflows, broadening the use of H&E imaging. 174 175 Systems such as Navigator can first locate regions of interest at low magnification and then examine them at higher magnification to characterize tumor and immune features, enabling efficient macro-to-micro-level assessment. 176 With the progress of AI, computational pathology for TME is advancing toward clinical application, providing greater support for personalized and precision medicine. Overall, these approaches improve analytical accuracy and efficiency, promote automation and standardization, and may accelerate clinical applications. Supplementary material online supplemental file 1 jitc-14-4-s001.pdf (394.7KB, pdf) DOI: 10.1136/jitc-2025-014429 online supplemental file 2 jitc-14-4-s002.pdf (1.4MB, pdf) DOI: 10.1136/jitc-2025-014429 Footnotes Funding: The authors acknowledge financial support by The National Natural Science Foundation of China (82574197), Healthy Zhejiang One Million People Cohort (K20230085), Key Research and Development Program of Zhejiang Province (2020C03002), The Leading Innovative and Entrepreneur Team Introduction Program of Zhejiang (2019R01007), and faculty startup funds. Provenance and peer review: Not commissioned; externally peer reviewed. Patient consent for publication: Not applicable. Ethics approval: Not applicable. References 1. Glaviano A, Lau HS-H, Carter LM, et al. Harnessing the tumor microenvironment: targeted cancer therapies through modulation of epithelial-mesenchymal transition. J Hematol Oncol . 2025;18:6. doi: 10.1186/s13045-024-01634-6. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 2. Wang L, Zhang L, Zhang Z, et al. Advances in targeting tumor microenvironment for immunotherapy. Front Immunol. 2024;15:1472772. doi: 10.3389/fimmu.2024.1472772. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 3. Van Bockstal MR, François A, Altinay S, et al. Interobserver variability in the assessment of stromal tumor-infiltrating lymphocytes (sTILs) in triple-negative invasive breast carcinoma influences the association with pathological complete response: the IVITA study. Mod Pathol. 2021;34:2130–40. doi: 10.1038/s41379-021-00865-z. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 4. Mardis ER. The Impact of Next-Generation Sequencing on Cancer Genomics: From Discovery to Clinic. Cold Spring Harb Perspect Med. 2019;9:a036269. doi: 10.1101/cshperspect.a036269. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Pentimalli TM, Karaiskos N, Rajewsky N. Challenges and Opportunities in the Clinical Translation of High-Resolution Spatial Transcriptomics. Annu Rev Pathol. 2025;20:405–32. doi: 10.1146/annurev-pathmechdis-111523-023417. [ DOI ] [ PubMed ] [ Google Scholar ] 6. Ren X, Zhang L, Zhang Y, et al. Insights Gained from Single-Cell Analysis of Immune Cells in the Tumor Microenvironment. Annu Rev Immunol. 2021;39:583–609. doi: 10.1146/annurev-immunol-110519-071134. [ DOI ] [ PubMed ] [ Google Scholar ] 7. Engblom C, Lundeberg J. Putting cancer immunotherapy into spatial context in the clinic. Nat Biotechnol. 2025;43:471–6. doi: 10.1038/s41587-025-02596-x. [ DOI ] [ PubMed ] [ Google Scholar ] 8. Verghese G, Lennerz JK, Ruta D, et al. Computational pathology in cancer diagnosis, prognosis, and prediction – present day and prospects. J Pathol. 2023;260:551–63. doi: 10.1002/path.6163. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 9. Bandi P, Geessink O, Manson Q, et al. From Detection of Individual Metastases to Classification of Lymph Node Status at the Patient Level: The CAMELYON17 Challenge. IEEE Trans Med Imaging. 2019;38:550–60. doi: 10.1109/TMI.2018.2867350. [ DOI ] [ PubMed ] [ Google Scholar ] 10. Ehteshami Bejnordi B, Veta M, Johannes van Diest P, et al. Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer. JAMA. 2017;318:2199. doi: 10.1001/jama.2017.14585. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. U.S. Food and Drug Administration . FDA; 2024. [12-May-2025]. FDA allows marketing of first whole slide imaging system for digital pathology. https://www.fda.gov/news-events/press-announcements/fda-allows-marketing-first-whole-slide-imaging-system-digital-pathology Available. Accessed. [ Google Scholar ] 12. Laurinavicius A, Rasmusson A, Plancoulaine B, et al. Machine-Learning-Based Evaluation of Intratumoral Heterogeneity and Tumor-Stroma Interface for Clinical Guidance. Am J Pathol. 2021;191:1724–31. doi: 10.1016/j.ajpath.2021.04.008. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 13. Basavanhally AN, Ganesan S, Agner S, et al. Computerized image-based detection and grading of lymphocytic infiltration in HER2+ breast cancer histopathology. IEEE Trans Biomed Eng. 2010;57:642–53. doi: 10.1109/TBME.2009.2035305. [ DOI ] [ PubMed ] [ Google Scholar ] 14. Setiadi AF, Ray NC, Kohrt HE, et al. Quantitative, architectural analysis of immune cell subsets in tumor-draining lymph nodes from breast cancer patients and healthy lymph nodes. PLoS One. 2010;5:e12420. doi: 10.1371/journal.pone.0012420. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 15. Rizzardi AE, Vogel RI, Koopmeiners JS, et al. Elevated hyaluronan and hyaluronan-mediated motility receptor are associated with biochemical failure in patients with intermediate-grade prostate tumors. Cancer. 2014;120:1800–9. doi: 10.1002/cncr.28646. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Lewis JS, Ali S, Luo J, et al. A Quantitative Histomorphometric Classifier (QuHbIC) Identifies Aggressive Versus Indolent p16-Positive Oropharyngeal Squamous Cell Carcinoma. Am J Surg Pathol. 2014;38:128–37. doi: 10.1097/PAS.0000000000000086. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 17. Heindl A, Nawaz S, Yuan Y. Mapping spatial heterogeneity in the tumor microenvironment: a new era for digital pathology. Laboratory Investigation . 2015;95:377–84. doi: 10.1038/labinvest.2014.155. [ DOI ] [ PubMed ] [ Google Scholar ] 18. Xu Y, Su G-H, Ma D, et al. Technological advances in cancer immunity: from immunogenomics to single-cell analysis and artificial intelligence. Sig Transduct Target Ther. 2021;6:312. doi: 10.1038/s41392-021-00729-7. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 19. Grabe N, Roth W, Foersch S. Digital pathology in immuno-oncology-current opportunities and challenges: overview of the analysis of immune cell infiltrates using whole slide imaging. Pathol. 2018;39:539–45. doi: 10.1007/s00292-018-0540-9. [ DOI ] [ PubMed ] [ Google Scholar ] 20. Beck AH, Sangoi AR, Leung S, et al. Systematic Analysis of Breast Cancer Morphology Uncovers Stromal Features Associated with Survival. Sci Transl Med. 2011;3 doi: 10.1126/scitranslmed.3002564. [ DOI ] [ PubMed ] [ Google Scholar ] 21. Yuan Y, Failmezger H, Rueda OM, et al. Quantitative image analysis of cellular heterogeneity in breast tumors complements genomic profiling. Sci Transl Med. 2012;4:157ra143. doi: 10.1126/scitranslmed.3004330. [ DOI ] [ PubMed ] [ Google Scholar ] 22. Graham S, Vu QD, Jahanifar M, et al. CoNIC Challenge: Pushing the frontiers of nuclear detection, segmentation, classification and counting. Med Image Anal. 2024;92:103047. doi: 10.1016/j.media.2023.103047. [ DOI ] [ PubMed ] [ Google Scholar ] 23. Gamper J, Koohbanani NA, Benes K, et al. PanNuke dataset extension, insights and baselines. arXiv. 2020 doi: 10.48550/arXiv.2003.10778. [ DOI ] [ Google Scholar ] 24. Kumar N, Verma R, Anand D, et al. A Multi-Organ Nucleus Segmentation Challenge. IEEE Trans Med Imaging. 2020;39:1380–91. doi: 10.1109/TMI.2019.2947628. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 25. Verma R, Kumar N, Patil A, et al. MoNuSAC2020: A Multi-Organ Nuclei Segmentation and Classification Challenge. IEEE Trans Med Imaging. 2021;40:3413–23. doi: 10.1109/TMI.2021.3085712. [ DOI ] [ PubMed ] [ Google Scholar ] 26. Pachitariu M, Rariden M, Stringer C. Cellpose-SAM: superhuman generalization for cellular segmentation. Bioinformatics. 2025 doi: 10.1101/2025.04.28.651001. Preprint. [ DOI ] 27. Hörst F, Rempe M, Heine L, et al. CellViT: Vision Transformers for precise cell segmentation and classification. Med Image Anal. 2024;94:103143. doi: 10.1016/j.media.2024.103143. [ DOI ] [ PubMed ] [ Google Scholar ] 28. Stringer C, Pachitariu M. Cellpose3: one-click image restoration for improved cellular segmentation. Nat Methods. 2025;22:592–9. doi: 10.1038/s41592-025-02595-5. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 29. Graham S, Vu QD, Raza SEA, et al. Hover-Net: Simultaneous segmentation and classification of nuclei in multi-tissue histology images. Med Image Anal. 2019;58:101563. doi: 10.1016/j.media.2019.101563. [ DOI ] [ PubMed ] [ Google Scholar ] 30. Weigert M, Schmidt U. Nuclei Instance Segmentation and Classification in Histopathology Images with Stardist. 2022 IEEE International Symposium on Biomedical Imaging Challenges (ISBIC) doi: 10.1109/ISBIC56247.2022.9854534. n.d. [ DOI ] [ Google Scholar ] 31. Javed S, Mahmood A, Fraz MM, et al. Cellular community detection for tissue phenotyping in colorectal cancer histology images. Med Image Anal. 2020;63:101696. doi: 10.1016/j.media.2020.101696. [ DOI ] [ PubMed ] [ Google Scholar ] 32. Isnaldi E, Richard F, De Schepper M, et al. Digital analysis of distant and cancer-associated mammary adipocytes. Breast. 2020;54:179–86. doi: 10.1016/j.breast.2020.10.004. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 33. Li Z, Mirjahanmardi SH, Sali R, et al. Automated cell annotation and classification on histopathology for spatial biomarker discovery. Nat Commun. 2025;16:6240. doi: 10.1038/s41467-025-61349-1. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 34. Lou S, Ji J, Li H, et al. A large histological images dataset of gastric cancer with tumour microenvironment annotation for AI. Sci Data. 2025;12:138. doi: 10.1038/s41597-025-04489-9. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 35. Kather JN, Halama N, Marx A. 100,000 histological images of human colorectal cancer and healthy tissue. 2018 36. Xu J, Luo X, Wang G, et al. A Deep Convolutional Neural Network for segmenting and classifying epithelial and stromal regions in histopathological images. Neurocomputing (Amst) 2016;191:214–23. doi: 10.1016/j.neucom.2016.01.034. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 37. Kludt C, Wang Y, Ahmad W, et al. Next-generation lung cancer pathology: Development and validation of diagnostic and prognostic algorithms. Cell Rep Med . 2024;5:101697. doi: 10.1016/j.xcrm.2024.101697. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 38. Lin S, Yong J, Zhang L, et al. Applying image features of proximal paracancerous tissues in predicting prognosis of patients with hepatocellular carcinoma. Comput Biol Med. 2024;173:108365. doi: 10.1016/j.compbiomed.2024.108365. [ DOI ] [ PubMed ] [ Google Scholar ] 39. Aggarwal A, Khalighi S, Babu D, et al. Computational pathology identifies immune-mediated collagen disruption to predict clinical outcomes in gynecologic malignancies. Commun Med (Lond) 2024;4:2. doi: 10.1038/s43856-023-00428-0. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 40. Zhang H, Chen L, Li L, et al. Prediction and analysis of tumor infiltrating lymphocytes across 28 cancers by TILScout using deep learning. NPJ Precis Oncol . 2025;9:76. doi: 10.1038/s41698-025-00866-0. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 41. Aggarwal A, Corredor G, Fu P, et al. Image analysis Uncovers associations between immune landscape, collagen structure, and neoadjuvant chemotherapy in high-grade serous ovarian carcinomas. Heliyon. 2024;10:e33618. doi: 10.1016/j.heliyon.2024.e33618. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 42. Gonzàlez-Farré M, Gibert J, Santiago-Díaz P, et al. Automated quantification of stromal tumour infiltrating lymphocytes is associated with prognosis in breast cancer. Virchows Arch . 2023;483:655–63. doi: 10.1007/s00428-023-03608-4. [ DOI ] [ PubMed ] [ Google Scholar ] 43. Thagaard J, Stovgaard ES, Vognsen LG, et al. Automated Quantification of sTIL Density with H&E-Based Digital Image Analysis Has Prognostic Potential in Triple-Negative Breast Cancers. Cancers (Basel) 2021;13:3050. doi: 10.3390/cancers13123050. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 44. Bose P, Brockton NT, Guggisberg K, et al. Fractal analysis of nuclear histology integrates tumor and stromal features into a single prognostic factor of the oral cancer microenvironment. BMC Cancer. 2015;15:409. doi: 10.1186/s12885-015-1380-0. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 45. Turkki R, Linder N, Kovanen PE, et al. Antibody-supervised deep learning for quantification of tumor-infiltrating immune cells in hematoxylin and eosin stained breast cancer samples. J Pathol Inform. 2016;7:38. doi: 10.4103/2153-3539.189703. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 46. Failmezger H, Muralidhar S, Rullan A, et al. Topological Tumor Graphs: A Graph-Based Spatial Model to Infer Stromal Recruitment for Immunosuppression in Melanoma Histology. Cancer Res. 2020;80:1199–209. doi: 10.1158/0008-5472.CAN-19-2268. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 47. Cheng J, Mo X, Wang X, et al. Identification of topological features in renal tumor microenvironment associated with patient survival. Bioinformatics. 2018;34:1024–30. doi: 10.1093/bioinformatics/btx723. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 48. Gallo E, Guardiani D, Betti M, et al. AI drives the assessment of lung cancer microenvironment composition. J Pathol Inform. 2024;15:100400. doi: 10.1016/j.jpi.2024.100400. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 49. Wang S, Wang T, Yang L, et al. ConvPath: A software tool for lung adenocarcinoma digital pathological image analysis aided by a convolutional neural network. EBioMedicine. 2019;50:103–10. doi: 10.1016/j.ebiom.2019.10.033. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 50. Wang H, Jiang Y, Li B, et al. Single-Cell Spatial Analysis of Tumor and Immune Microenvironment on Whole-Slide Image Reveals Hepatocellular Carcinoma Subtypes. Cancers (Basel) 2020;12:3562. doi: 10.3390/cancers12123562. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 51. Lloyd MC, Alfarouk KO, Verduzco D, et al. Vascular measurements correlate with estrogen receptor status. BMC Cancer. 2014;14:279. doi: 10.1186/1471-2407-14-279. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 52. Feng Z, Lin H, Liu Z, et al. Artificial intelligence-quantified tumour-lymphocyte spatial interaction predicts disease-free survival in resected lung adenocarcinoma: A graph-based, multicentre study. Comput Methods Programs Biomed. 2023;238:107617. doi: 10.1016/j.cmpb.2023.107617. [ DOI ] [ PubMed ] [ Google Scholar ] 53. Aswolinskiy W, Munari E, Horlings HM, et al. PROACTING: predicting pathological complete response to neoadjuvant chemotherapy in breast cancer from routine diagnostic histopathology biopsies with deep learning. Breast Cancer Res . 2023;25:142. doi: 10.1186/s13058-023-01726-0. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 54. Balancin ML, Teodoro WR, Farhat C, et al. An integrative histopathologic clustering model based on immuno-matrix elements to predict the risk of death in malignant mesothelioma. Cancer Med. 2020;9:4836–49. doi: 10.1002/cam4.3111. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 55. Narayanan PL, Raza SEA, Hall AH, et al. Unmasking the immune microecology of ductal carcinoma in situ with deep learning. NPJ Breast Cancer. 2021;7:19. doi: 10.1038/s41523-020-00205-5. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 56. Zhou S, Xie Y, Feng X, et al. Artificial intelligence in gastrointestinal cancer research: Image learning advances and applications. Cancer Lett. 2025;614:217555. doi: 10.1016/j.canlet.2025.217555. [ DOI ] [ PubMed ] [ Google Scholar ] 57. Zhao K, Li Z, Yao S, et al. Artificial intelligence quantified tumour-stroma ratio is an independent predictor for overall survival in resectable colorectal cancer. EBioMedicine. 2020;61:103054. doi: 10.1016/j.ebiom.2020.103054. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 58. Lu Z, Zhan X, Wu Y, et al. BrcaSeg: A Deep Learning Approach for Tissue Quantification and Genomic Correlations of Histopathological Images. Genomics Proteomics Bioinformatics. 2021;19:1032–42. doi: 10.1016/j.gpb.2020.06.026. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 59. Jiao Y, Li J, Qian C, et al. Deep learning-based tumor microenvironment analysis in colon adenocarcinoma histopathological whole-slide images. Comput Methods Programs Biomed. 2021;204:106047. doi: 10.1016/j.cmpb.2021.106047. [ DOI ] [ PubMed ] [ Google Scholar ] 60. Jones HJS, Cunningham C, Askautrud HA, et al. Stromal composition predicts recurrence of early rectal cancer after local excision. Histopathology. 2021;79:947–56. doi: 10.1111/his.14438. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 61. Jakab A, Patai ÁV, Micsik T. Digital image analysis provides robust tissue microenvironment-based prognosticators in patients with stage I-IV colorectal cancer. Hum Pathol. 2022;128:141–51. doi: 10.1016/j.humpath.2022.07.003. [ DOI ] [ PubMed ] [ Google Scholar ] 62. Brummel K, Eerkens AL, de Bruyn M, et al. Tumour-infiltrating lymphocytes: from prognosis to treatment selection. Br J Cancer. 2023;128:451–8. doi: 10.1038/s41416-022-02119-4. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 63. Noy R, Pollard JW. Tumor-associated macrophages: from mechanisms to therapy. Immunity. 2014;41:49–61. doi: 10.1016/j.immuni.2014.06.010. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 64. Saltz J, Gupta R, Hou L, et al. Spatial Organization and Molecular Correlation of Tumor-Infiltrating Lymphocytes Using Deep Learning on Pathology Images. Cell Rep. 2018;23:181–93. doi: 10.1016/j.celrep.2018.03.086. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 65. Park S, Ock C-Y, Kim H, et al. Artificial Intelligence–Powered Spatial Analysis of Tumor-Infiltrating Lymphocytes as Complementary Biomarker for Immune Checkpoint Inhibition in Non–Small-Cell Lung Cancer. JCO . 2022;40:1916–28. doi: 10.1200/JCO.21.02010. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 66. Eddy JA, Thorsson V, Lamb AE, et al. CRI iAtlas: an interactive portal for immuno-oncology research. F1000Res. 2020;9:1028. doi: 10.12688/f1000research.25141.1. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 67. Harmon SA, Sanford TH, Brown GT, et al. Multiresolution Application of Artificial Intelligence in Digital Pathology for Prediction of Positive Lymph Nodes From Primary Tumors in Bladder Cancer. JCO Clin Cancer Inform. 2020;4:367–82. doi: 10.1200/CCI.19.00155. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 68. Choi S, Cho SI, Jung W, et al. Deep learning model improves tumor-infiltrating lymphocyte evaluation and therapeutic response prediction in breast cancer. NPJ Breast Cancer. 2023;9:71. doi: 10.1038/s41523-023-00577-4. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 69. Kim JY, Jeong H, Valero Puche A, et al. Artificial Intelligence-Driven Quantification of Tumor-Stroma Ratio and Fibroblasts Enables Precise Classification of Stroma Quality and Quantity in Predicting Colorectal Cancer Recurrence. Pathobiology . 2025;92:276–87. doi: 10.1159/000546021. [ DOI ] [ PubMed ] [ Google Scholar ] 70. Shen J, Choi Y-L, Lee T, et al. Inflamed immune phenotype predicts favorable clinical outcomes of immune checkpoint inhibitor therapy across multiple cancer types. J Immunother Cancer. 2024;12:e008339. doi: 10.1136/jitc-2023-008339. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 71. Jung HA, Park K-U, Cho S, et al. A Phase II Study of Nivolumab plus Gemcitabine in Patients with Recurrent or Metastatic Nasopharyngeal Carcinoma (KCSG HN17–11) Clin Cancer Res. 2022;28:4240–7. doi: 10.1158/1078-0432.CCR-22-1238. [ DOI ] [ PubMed ] [ Google Scholar ] 72. Diao JA, Wang JK, Chui WF, et al. Human-interpretable image features derived from densely mapped cancer pathology slides predict diverse molecular phenotypes. Nat Commun. 2021;12:1613. doi: 10.1038/s41467-021-21896-9. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 73. Aggarwal A, Jana M, Singh A, et al. Artificial intelligence-based virtual staining platform for identifying tumor-associated macrophages from hematoxylin and eosin-stained images. Eur J Cancer. 2025;220:115390. doi: 10.1016/j.ejca.2025.115390. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 74. van Rijthoven M, Obahor S, Pagliarulo F, et al. Multi-resolution deep learning characterizes tertiary lymphoid structures and their prognostic relevance in solid tumors. Commun Med (Lond) 2024;4:5. doi: 10.1038/s43856-023-00421-7. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 75. Chen Z, Wang X, Jin Z, et al. Deep learning on tertiary lymphoid structures in hematoxylin-eosin predicts cancer prognosis and immunotherapy response. NPJ Precis Oncol . 2024;8:73. doi: 10.1038/s41698-024-00579-w. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 76. Datta RR, Schran S, Persa O-D, et al. Post-transplant Malignancies Show Reduced T-cell Abundance and Tertiary Lymphoid Structures as Correlates of Impaired Cancer Immunosurveillance. Clin Cancer Res. 2022;28:1712–23. doi: 10.1158/1078-0432.CCR-21-3746. [ DOI ] [ PubMed ] [ Google Scholar ] 77. Barmpoutis P, Di Capite M, Kayhanian H, et al. Tertiary lymphoid structures (TLS) identification and density assessment on H&E-stained digital slides of lung cancer. PLoS One. 2021;16:e0256907. doi: 10.1371/journal.pone.0256907. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 78. Wang X, Tao R, Li T, et al. 137 Tertiary lymphoid structure signature in digital H&E slides as a prognostic biomarker in cutaneous melanoma. SITC 39th Annual Meeting (SITC 2024) Abstracts; 2024. [ DOI ] [ Google Scholar ] 79. Saqi A, Liu Y, Politis MG, et al. Combined expert-in-the-loop—random forest multiclass segmentation U-net based artificial intelligence model: evaluation of non-small cell lung cancer in fibrotic and non-fibrotic microenvironments. J Transl Med. 2024;22:640. doi: 10.1186/s12967-024-05394-2. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 80. Shi L, Zhang Y, Wang H. Prognostic prediction based on histopathologic features of tumor microenvironment in colorectal cancer. Front Med. 2023;10:1154077. doi: 10.3389/fmed.2023.1154077. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 81. Galon J, Costes A, Sanchez-Cabo F, et al. Type, Density, and Location of Immune Cells Within Human Colorectal Tumors Predict Clinical Outcome. Science. 2006;313:1960–4. doi: 10.1126/science.1129139. [ DOI ] [ PubMed ] [ Google Scholar ] 82. Mahmoud SMA, Paish EC, Powe DG, et al. Tumor-infiltrating CD8+ lymphocytes predict clinical outcome in breast cancer. JCO. 2011;29:1949–55. doi: 10.1200/JCO.2010.30.5037. [ DOI ] [ PubMed ] [ Google Scholar ] 83. Wang S, Rong R, Zhou Q, et al. Deep learning of cell spatial organizations identifies clinically relevant insights in tissue images. Nat Commun. 2023;14:7872. doi: 10.1038/s41467-023-43172-8. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 84. Xie J, Pu X, He J, et al. Survival prediction on intrahepatic cholangiocarcinoma with histomorphological analysis on the whole slide images. Comput Biol Med. 2022;146:105520. doi: 10.1016/j.compbiomed.2022.105520. [ DOI ] [ PubMed ] [ Google Scholar ] 85. Shaban M, Raza SEA, Hassan M, et al. A digital score of tumour‐associated stroma infiltrating lymphocytes predicts survival in head and neck squamous cell carcinoma. J Pathol. 2022;256:174–85. doi: 10.1002/path.5819. [ DOI ] [ PubMed ] [ Google Scholar ] 86. Barroso VM, Weng Z, Glamann L, et al. Artificial Intelligence–Based Single-Cell Analysis as a Next-Generation Histologic Grading Approach in Colorectal Cancer: Prognostic Role and Tumor Biology Assessment. Mod Pathol. 2025;38:100771. doi: 10.1016/j.modpat.2025.100771. [ DOI ] [ PubMed ] [ Google Scholar ] 87. Rączkowska A, Paśnik I, Kukiełka M, et al. Deep learning-based tumor microenvironment segmentation is predictive of tumor mutations and patient survival in non-small-cell lung cancer. BMC Cancer. 2022;22:1001. doi: 10.1186/s12885-022-10081-w. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 88. Surace M, Rognoni L, Rodriguez-Canales J. Characterization of the immune microenvironment of NSCLC by multispectral analysis of multiplex immunofluorescence images. Methods Enzymol. 2020;635:33–50. doi: 10.1016/bs.mie.2019.07.039. [ DOI ] [ PubMed ] [ Google Scholar ] 89. Wang X, Barrera C, Bera K, et al. Spatial interplay patterns of cancer nuclei and tumor-infiltrating lymphocytes (TILs) predict clinical benefit for immune checkpoint inhibitors. Sci Adv. 2022;8:eabn3966. doi: 10.1126/sciadv.abn3966. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 90. Taube JM, Sunshine JC, Angelo M, et al. Society for Immunotherapy of Cancer: updates and best practices for multiplex immunohistochemistry (IHC) and immunofluorescence (IF) image analysis and data sharing. J Immunother Cancer. 2025;13:e008875. doi: 10.1136/jitc-2024-008875. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 91. Gong C, Anders RA, Zhu Q, et al. Quantitative characterization of CD8+ T cell clustering and spatial heterogeneity in solid tumors. Front Oncol. 2018;8:649. doi: 10.3389/fonc.2018.00649. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 92. Kather JN, Suarez-Carmona M, Charoentong P, et al. Topography of cancer-associated immune cells in human solid tumors. Elife. 2018;7:e36967. doi: 10.7554/eLife.36967. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 93. Desbois M, Udyavar AR, Ryner L, et al. Integrated digital pathology and transcriptome analysis identifies molecular mediators of T-cell exclusion in ovarian cancer. Nat Commun. 2020;11:5583. doi: 10.1038/s41467-020-19408-2. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 94. Mi H, Gong C, Sulam J, et al. Digital Pathology Analysis Quantifies Spatial Heterogeneity of CD3, CD4, CD8, CD20, and FoxP3 Immune Markers in Triple-Negative Breast Cancer. Front Physiol. 11 doi: 10.3389/fphys.2020.583333. n.d. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 95. Klughammer J, Kiesel B, Roetzer T, et al. The DNA methylation landscape of glioblastoma disease progression shows extensive heterogeneity in time and space. Nat Med. 2018;24:1611–24. doi: 10.1038/s41591-018-0156-x. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 96. Zwing N, Failmezger H, Ooi C-H, et al. Analysis of Spatial Organization of Suppressive Myeloid Cells and Effector T Cells in Colorectal Cancer-A Potential Tool for Discovering Prognostic Biomarkers in Clinical Research. Front Immunol. 2020;11:550250. doi: 10.3389/fimmu.2020.550250. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 97. Ono A, Terada Y, Kawata T, et al. Assessment of associations between clinical and immune microenvironmental factors and tumor mutation burden in resected nonsmall cell lung cancer by applying machine learning to whole-slide images. Cancer Med. 2020;9:4864–75. doi: 10.1002/cam4.3107. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 98. de Haan K, Zhang Y, Zuckerman JE, et al. Deep learning-based transformation of H&E stained tissues into special stains. Nat Commun. 2021;12:4884. doi: 10.1038/s41467-021-25221-2. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 99. Liu S, Zhu C, Xu F, et al. BCI: breast cancer immunohistochemical image generation through pyramid pix2pix. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW); New Orleans, LA, USA. pp. 1814–23. [ DOI ] [ Google Scholar ] 100. Zhou Z, Jiang Y, Sun Z, et al. Virtual multiplexed immunofluorescence staining from non-antibody-stained fluorescence imaging for gastric cancer prognosis. EBioMedicine. 2024;107:105287. doi: 10.1016/j.ebiom.2024.105287. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 101. Chen F, Zhang R, Zheng B, et al. In: Medical image computing and computer assisted intervention – MICCAI 2024. Linguraru MG, Dou Q, Feragen A, editors. Cham: Springer Nature Switzerland; 2024. Pathological semantics-preserving learning for H&E-to-IHC virtual staining; pp. 384–94. [ Google Scholar ] 102. Andani S, Chen B, Ficek-Pascual J, et al. Histopathology-based protein multiplex generation using deep learning. Nat Mach Intell . 2025;7:1292–307. doi: 10.1038/s42256-025-01074-y. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 103. Klöckner P, Teixeira J, Montezuma D, et al. H&E to IHC virtual staining methods in breast cancer: an overview and benchmarking. npj Digit Med. 2025;8:384. doi: 10.1038/s41746-025-01741-9. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 104. Valanarasu JMJ, Xu H, Usuyama N, et al. Multimodal AI generates virtual population for tumor microenvironment modeling. Cell. 2026;189:386–400. doi: 10.1016/j.cell.2025.11.016. [ DOI ] [ PubMed ] [ Google Scholar ] 105. Pati P, Karkampouna S, Bonollo F, et al. Accelerating histopathology workflows with generative AI-based virtually multiplexed tumour profiling. Nat Mach Intell . 2024;6:1077–93. doi: 10.1038/s42256-024-00889-5. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 106. Wu E, Bieniosek M, Wu Z, et al. ROSIE: AI generation of multiplex immunofluorescence staining from histopathology images. Nat Commun. 2025;16:7633. doi: 10.1038/s41467-025-62346-0. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 107. Bai B, Yang X, Li Y, et al. Deep learning-enabled virtual histological staining of biological samples. Light Sci Appl. 2023;12:57. doi: 10.1038/s41377-023-01104-7. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 108. Brattoli B, Mostafavi M, Lee T, et al. A universal immunohistochemistry analyzer for generalizing AI-driven assessment of immunohistochemistry across immunostains and cancer types. NPJ Precis Oncol . 2024;8:277. doi: 10.1038/s41698-024-00770-z. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 109. Hornburg M, Desbois M, Lu S, et al. Single-cell dissection of cellular components and interactions shaping the tumor immune phenotypes in ovarian cancer. Cancer Cell. 2021;39:928–44. doi: 10.1016/j.ccell.2021.04.004. [ DOI ] [ PubMed ] [ Google Scholar ] 110. Grünwald BT, Devisme A, Andrieux G, et al. Spatially confined sub-tumor microenvironments in pancreatic cancer. Cell. 2021;184:5577–92. doi: 10.1016/j.cell.2021.09.022. [ DOI ] [ PubMed ] [ Google Scholar ] 111. Hue J, Valinciute Z, Thavaraj S, et al. Multifactorial estimation of clinical outcome in HPV-associated oropharyngeal squamous cell carcinoma via automated image analysis of routine diagnostic H&E slides and neural network modelling. Oral Oncol. 2023;141:106399. doi: 10.1016/j.oraloncology.2023.106399. [ DOI ] [ PubMed ] [ Google Scholar ] 112. Liu B, Polack M, Coudray N, et al. Self-supervised learning reveals clinically relevant histomorphological patterns for therapeutic strategies in colon cancer. Nat Commun. 2025;16:2328. doi: 10.1038/s41467-025-57541-y. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 113. Li Y, Dou Y, Da Veiga Leprevost F, et al. Proteogenomic data and resources for pan-cancer analysis. Cancer Cell. 2023;41:1397–406. doi: 10.1016/j.ccell.2023.06.009. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 114. Weinstein JN, Collisson EA, Mills GB, et al. The Cancer Genome Atlas Pan-Cancer analysis project. Nat Genet. 2013;45:1113–20. doi: 10.1038/ng.2764. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 115. Zhan X, Cheng J, Huang Z, et al. Correlation Analysis of Histopathology and Proteogenomics Data for Breast Cancer. Mol Cell Proteomics. 2019;18:S37–51. doi: 10.1074/mcp.RA118.001232. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 116. Chang H, Fontenay GV, Han J, et al. Morphometic analysis of TCGA glioblastoma multiforme. BMC Bioinformatics. 2011;12:484. doi: 10.1186/1471-2105-12-484. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 117. Kong J, Cooper LAD, Wang F, et al. Machine-based morphologic analysis of glioblastoma using whole-slide pathology images uncovers clinically relevant molecular correlates. PLoS ONE. 2013;8:e81049. doi: 10.1371/journal.pone.0081049. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 118. Wang C, Pécot T, Zynger DL, et al. Identifying survival associated morphological features of triple negative breast cancer using multiple datasets. J Am Med Inform Assoc. 2013;20:680–7. doi: 10.1136/amiajnl-2012-001538. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 119. Cooper LAD, Kong J, Gutman DA, et al. Integrated morphologic analysis for the identification and characterization of disease subtypes. J Am Med Inform Assoc . 2012;19:317–23. doi: 10.1136/amiajnl-2011-000700. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 120. Azher ZL, Suvarna A, Chen J-Q, et al. Assessment of emerging pretraining strategies in interpretable multimodal deep learning for cancer prognostication. BioData Min. 2023;16:23. doi: 10.1186/s13040-023-00338-w. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 121. Ding K, Zhou M, Wang H, et al. Spatially aware graph neural networks and cross-level molecular profile prediction in colon cancer histopathology: a retrospective multi-cohort study. Lancet Digit Health. 2022;4:e787–95. doi: 10.1016/S2589-7500(22)00168-6. [ DOI ] [ PubMed ] [ Google Scholar ] 122. Chen Z, Li X, Yang M, et al. Optimization of deep learning models for the prediction of gene mutations using unsupervised clustering. The Journal of Pathology CR. 2023;9:3–17. doi: 10.1002/cjp2.302. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 123. Fisher TB, Saini G, Rekha TS, et al. Digital image analysis and machine learning-assisted prediction of neoadjuvant chemotherapy response in triple-negative breast cancer. Breast Cancer Res . 2024;26:12. doi: 10.1186/s13058-023-01752-y. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 124. Zheng X, Jing B, Zhao Z, et al. An interpretable deep learning model for identifying the morphological characteristics of dMMR/MSI-H gastric cancer. iScience. 2024;27:109243. doi: 10.1016/j.isci.2024.109243. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 125. Kather JN, Pearson AT, Halama N, et al. Deep learning can predict microsatellite instability directly from histology in gastrointestinal cancer. Nat Med. 2019;25:1054–6. doi: 10.1038/s41591-019-0462-y. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 126. Muti HS, Heij LR, Keller G, et al. Development and validation of deep learning classifiers to detect Epstein-Barr virus and microsatellite instability status in gastric cancer: a retrospective multicentre cohort study. Lancet Digit Health. 2021;3:e654–64. doi: 10.1016/S2589-7500(21)00133-3. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 127. Saillard C, Dubois R, Tchita O, et al. Validation of MSIntuit as an AI-based pre-screening tool for MSI detection from colorectal cancer histology slides. Nat Commun. 2023;14:6695. doi: 10.1038/s41467-023-42453-6. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 128. El Nahhas OSM, van Treeck M, Wölflein G, et al. From whole-slide image to biomarker prediction: end-to-end weakly supervised deep learning in computational pathology. Nat Protoc. 2025;20:293–316. doi: 10.1038/s41596-024-01047-2. [ DOI ] [ PubMed ] [ Google Scholar ] 129. Chen Y, Sun Z, Chen W, et al. The Immune Subtypes and Landscape of Gastric Cancer and to Predict Based on the Whole-Slide Images Using Deep Learning. Front Immunol. 2021;12:685992. doi: 10.3389/fimmu.2021.685992. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 130. Lapuente-Santana Ó, Kant J, Eduati F. Integrating histopathology and transcriptomics for spatial tumor microenvironment profiling in a melanoma case study. NPJ Precis Oncol. 2024;8:254. doi: 10.1038/s41698-024-00749-w. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 131. Liu X-P, Jin X, Seyed Ahmadian S, et al. Clinical significance and molecular annotation of cellular morphometric subtypes in lower-grade gliomas discovered by machine learning. Neuro-oncology. 2023;25:68–81. doi: 10.1093/neuonc/noac154. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 132. Jiang Y, Huang S, Zhu X, et al. Artificial Intelligence Meets Whole Slide Images: Deep Learning Model Shapes an Immune-Hot Tumor and Guides Precision Therapy in Bladder Cancer. J Oncol. 2022;2022:1–27. doi: 10.1155/2022/8213321. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 133. Qaiser T, Lee C-Y, Vandenberghe M, et al. Usability of deep learning and H&E images predict disease outcome-emerging tool to optimize clinical trials. NPJ Precis Onc. 2022;6:37. doi: 10.1038/s41698-022-00275-7. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 134. Bilal M, Tsang YW, Ali M, et al. Development and validation of artificial intelligence-based prescreening of large-bowel biopsies taken in the UK and Portugal: a retrospective cohort study. Lancet Digit Health. 2023;5:e786–97. doi: 10.1016/S2589-7500(23)00148-6. [ DOI ] [ PubMed ] [ Google Scholar ] 135. Wang R, Khurram SA, Walsh H, et al. A Novel Deep Learning Algorithm for Human Papillomavirus Infection Prediction in Head and Neck Cancers Using Routine Histology Images. Mod Pathol. 2023;36:100320. doi: 10.1016/j.modpat.2023.100320. [ DOI ] [ PubMed ] [ Google Scholar ] 136. Wu X, Yan H, Qiu M, et al. Comprehensive characterization of tumor microenvironment in colorectal cancer via molecular analysis. Elife. 2023;12 doi: 10.7554/eLife.86032. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 137. Manescu P, Geradts J, Fernandez-Reyes D. Computational Pathology Detection of Hypoxia-Induced Morphologic Changes in Breast Cancer. Am J Pathol. 2025;195:663–70. doi: 10.1016/j.ajpath.2024.10.023. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 138. Wang C-W, Liu T-C, Lai P-J, et al. Ensemble transformer-based multiple instance learning to predict pathological subtypes and tumor mutational burden from histopathological whole slide images of endometrial and colorectal cancer. Med Image Anal. 2025;99:103372. doi: 10.1016/j.media.2024.103372. [ DOI ] [ PubMed ] [ Google Scholar ] 139. El Nahhas OSM, Loeffler CML, Carrero ZI, et al. Regression-based Deep-Learning predicts molecular biomarkers from pathology slides. Nat Commun. 2024;15:1253. doi: 10.1038/s41467-024-45589-1. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 140. Xu H, Usuyama N, Bagga J, et al. A whole-slide foundation model for digital pathology from real-world data. Nature New Biol. 2024;630:181–8. doi: 10.1038/s41586-024-07441-w. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 141. Wang X, Yang S, Zhang J, et al. Transformer-based unsupervised contrastive learning for histopathological image classification. Med Image Anal. 2022;81:102559. doi: 10.1016/j.media.2022.102559. [ DOI ] [ PubMed ] [ Google Scholar ] 142. Chen RJ, Ding T, Lu MY, et al. Towards a general-purpose foundation model for computational pathology. Nat Med. 2024;30:850–62. doi: 10.1038/s41591-024-02857-3. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 143. Zimmermann E, Vorontsov E, Viret J, et al. Virchow2: scaling self-supervised mixed magnification models in pathology. arXiv. 2024 doi: 10.48550/arXiv.2408.00738. [ DOI ] [ Google Scholar ] 144. Ning Z, Yang B, Wang Y, et al. Dual-path neural network extracts tumor microenvironment information from whole slide images to predict molecular typing and prognosis of Glioma. Comput Methods Programs Biomed. 2025;261:108580. doi: 10.1016/j.cmpb.2024.108580. [ DOI ] [ PubMed ] [ Google Scholar ] 145. Liu Y, Dai Y, Wang L. Spatial omics at the forefront: emerging technologies, analytical innovations, and clinical applications. Cancer Cell. 2026;44:24–49. doi: 10.1016/j.ccell.2025.12.009. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 146. Kulasinghe A, Wood F, Belz G. The seductive allure of spatial biology: accelerating new discoveries in the life sciences. Immunol Cell Biol. 2023;101:798–804. doi: 10.1111/imcb.12669. [ DOI ] [ PubMed ] [ Google Scholar ] 147. Tan X, Mulay O, Xie J, et al. Robust and interpretable prediction of gene markers and cell types from spatial transcriptomics data. Nat Commun. 2026;17:1781. doi: 10.1038/s41467-026-68487-0. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 148. Ge Y, Leng J, Tang Z, et al. Deep Learning-Enabled Integration of Histology and Transcriptomics for Tissue Spatial Profile Analysis. Research (Lambertville) 2025;8:568. doi: 10.34133/research.0568. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 149. Pham D, Tan X, Balderson B, et al. Robust mapping of spatiotemporal trajectories and cell-cell interactions in healthy and diseased tissues. Nat Commun. 2023;14:7739. doi: 10.1038/s41467-023-43120-6. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 150. Zhao E, Stone MR, Ren X, et al. Spatial transcriptomics at subspot resolution with BayesSpace. Nat Biotechnol. 2021;39:1375–84. doi: 10.1038/s41587-021-00935-2. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 151. Hu J, Li X, Coleman K, et al. SpaGCN: Integrating gene expression, spatial location and histology to identify spatial domains and spatially variable genes by graph convolutional network. Nat Methods. 2021;18:1342–51. doi: 10.1038/s41592-021-01255-8. [ DOI ] [ PubMed ] [ Google Scholar ] 152. Dong K, Zhang S. Deciphering spatial domains from spatially resolved transcriptomics with an adaptive graph attention auto-encoder. Nat Commun. 2022;13:1739. doi: 10.1038/s41467-022-29439-6. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 153. Chen W, Zhang P, Tran TN, et al. A visual–omics foundation model to bridge histopathology with spatial transcriptomics. Nat Methods. 2025;22:1568–82. doi: 10.1038/s41592-025-02707-1. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 154. Chelebian E, Avenel C, Wählby C. Combining spatial transcriptomics with tissue morphology. Nat Commun. 2025;16:4452. doi: 10.1038/s41467-025-58989-8. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 155. Reiman D, Sha L, Ho I, et al. Integrating rna expression and visual features for immune infiltrate prediction. Pac Symp Biocomput. 2019;24:284–95. [ PubMed ] [ Google Scholar ] 156. Hoang D-T, Dinstag G, Shulman ED, et al. A deep-learning framework to predict cancer treatment response from histopathology images through imputed transcriptomics. Nat Cancer . 2024;5:1305–17. doi: 10.1038/s43018-024-00793-2. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 157. Chen RJ, Lu MY, Wang J, et al. Pathomic Fusion: An Integrated Framework for Fusing Histopathology and Genomic Features for Cancer Diagnosis and Prognosis. IEEE Trans Med Imaging. 2022;41:757–70. doi: 10.1109/TMI.2020.3021387. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 158. Yu Y, Ouyang W, Huang Y, et al. Artificial intelligence-based multi-modal multi-tasks analysis reveals tumor molecular heterogeneity, predicts preoperative lymph node metastasis and prognosis in papillary thyroid carcinoma: a retrospective study. Int J Surg. 2025;111:839–56. doi: 10.1097/JS9.0000000000001875. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 159. Mao N, Dai Y, Zhou H, et al. A multimodal and fully automated system for prediction of pathological complete response to neoadjuvant chemotherapy in breast cancer. Sci Adv. 2025;11:eadr1576. doi: 10.1126/sciadv.adr1576. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 160. Wang Q, Zhao F, Zhang H, et al. Deep learning-based multi-task prediction of response to neoadjuvant chemotherapy using multiscale whole slide images in breast cancer: A multicenter study. Chin J Cancer Res. 2025;37:28–47. doi: 10.21147/j.issn.1000-9604.2025.01.03. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 161. Sammut S-J, Crispin-Ortuzar M, Chin S-F, et al. Multi-omic machine learning predictor of breast cancer therapy response. Nature New Biol. 2022;601:623–9. doi: 10.1038/s41586-021-04278-5. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 162. van der Leun AM, Thommen DS, Schumacher TN. CD8+ T cell states in human cancer: insights from single-cell analysis. Nat Rev Cancer. 2020;20:218–32. doi: 10.1038/s41568-019-0235-4. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 163. Asif A, Rajpoot K, Graham S, et al. Unleashing the potential of AI for pathology: challenges and recommendations. J Pathol. 2023;260:564–77. doi: 10.1002/path.6168. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 164. Benimam MM, Meas-Yedid V, Mukherjee S, et al. Statistical analysis of spatial patterns in tumor microenvironment images. Nat Commun. 2025;16:3090. doi: 10.1038/s41467-025-57943-y. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 165. Hong Y, Heo YJ, Kim B, et al. Deep learning-based virtual cytokeratin staining of gastric carcinomas to measure tumor–stroma ratio. Sci Rep. 2021;11:19255. doi: 10.1038/s41598-021-98857-1. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 166. Nikolaou N, Salazar D, RaviPrakash H, et al. A machine learning approach for multimodal data fusion for survival prediction in cancer patients. NPJ Precis Oncol . 2025;9:128. doi: 10.1038/s41698-025-00917-6. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 167. Mohr AE, Ortega-Santos CP, Whisner CM, et al. Navigating Challenges and Opportunities in Multi-Omics Integration for Personalized Healthcare. Biomedicines. 2024;12:1496. doi: 10.3390/biomedicines12071496. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 168. Liaqat MI, Nawaz S, Zaheer MZ, et al. Chameleon: A Multimodal Learning Framework Robust to Missing Modalities. Int J Multimed Info Retr. 2025;14:21. doi: 10.1007/s13735-025-00370-y. [ DOI ] [ Google Scholar ] 169. Weng Z, Seper A, Pryalukhin A, et al. GrandQC: A comprehensive solution to quality control problem in digital pathology. Nat Commun. 2024;15:10685. doi: 10.1038/s41467-024-54769-y. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 170. Janowczyk A, Zuo R, Gilmore H, et al. HistoQC: An Open-Source Quality Control Tool for Digital Pathology Slides. JCO Clin Cancer Inform. 2019;3:1–7. doi: 10.1200/CCI.18.00157. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 171. Zhang X-M, Gao T-H, Cai Q-Y, et al. Artificial intelligence in digital pathology diagnosis and analysis: technologies, challenges, and future prospects. Military Med Res. 2026;12:93. doi: 10.1186/s40779-025-00680-6. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 172. Richens JG, Lee CM, Johri S. Improving the accuracy of medical diagnosis with causal machine learning. Nat Commun. 2020;11:3923. doi: 10.1038/s41467-020-17419-7. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 173. Aggarwal A, Bharadwaj S, Corredor G, et al. Artificial intelligence in digital pathology - time for a reality check. Nat Rev Clin Oncol. 2025;22:283–91. doi: 10.1038/s41571-025-00991-6. [ DOI ] [ PubMed ] [ Google Scholar ] 174. Kovacheva VN, Snead D, Rajpoot NM. A model of the spatial tumour heterogeneity in colorectal adenocarcinoma tissue. BMC Bioinformatics. 2016;17:255. doi: 10.1186/s12859-016-1126-2. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 175. Song AH, Williams M, Williamson DFK, et al. Analysis of 3D pathology samples using weakly supervised AI. Cell. 2024;187:2502–20. doi: 10.1016/j.cell.2024.03.035. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 176. Tran M, Wagner S, Weichert W, et al. Navigating Through Whole Slide Images With Hierarchy, Multi-Object, and Multi-Scale Data. IEEE Trans Med Imaging. 2025;44:2002–15. doi: 10.1109/TMI.2025.3532728. [ DOI ] [ PubMed ] [ Google Scholar ] 177. Kather JN, Krisam J, Charoentong P, et al. Predicting survival from colorectal cancer histology slides using deep learning: A retrospective multicenter study. PLoS Med. 2019;16:e1002730. doi: 10.1371/journal.pmed.1002730. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 178. Baumann E, Dislich B, Rumberger JL, et al. HoVer-next: a fast nuclei segmentation and classification pipeline for next generation histopathology. Proceedings of The 7nd International Conference on Medical Imaging with Deep Learning, PMLR; 2024. [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials online supplemental file 1 jitc-14-4-s001.pdf (394.7KB, pdf) DOI: 10.1136/jitc-2025-014429 online supplemental file 2 jitc-14-4-s002.pdf (1.4MB, pdf) DOI: 10.1136/jitc-2025-014429 Articles from Journal for Immunotherapy of Cancer are provided here courtesy of BMJ Publishing Group ACTIONS View on publisher site PDF (5.2 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top