Radiomics features and clinical factors for predicting restenosis following endovascular therapy in patients with peripheral artery disease.

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 BMC Med Inform Decis Mak . 2026 Mar 4;26:118. doi: 10.1186/s12911-026-03421-4 Search in PMC Search in PubMed View in NLM Catalog Add to search Radiomics features and clinical factors for predicting restenosis following endovascular therapy in patients with peripheral artery disease Min Luo Min Luo 1 Department of Vascular Surgery, The First Affiliated Hospital of Kunming Medical University, 295 Xichang Rd, Kunming, 650000 China Find articles by Min Luo 1 , Haohua Wang Haohua Wang 1 Department of Vascular Surgery, The First Affiliated Hospital of Kunming Medical University, 295 Xichang Rd, Kunming, 650000 China Find articles by Haohua Wang 1 , Hai Xia Hai Xia 1 Department of Vascular Surgery, The First Affiliated Hospital of Kunming Medical University, 295 Xichang Rd, Kunming, 650000 China Find articles by Hai Xia 1 , Wangxing Feng Wangxing Feng 1 Department of Vascular Surgery, The First Affiliated Hospital of Kunming Medical University, 295 Xichang Rd, Kunming, 650000 China Find articles by Wangxing Feng 1 , Bin Li Bin Li 1 Department of Vascular Surgery, The First Affiliated Hospital of Kunming Medical University, 295 Xichang Rd, Kunming, 650000 China Find articles by Bin Li 1 , Yong Li Yong Li 1 Department of Vascular Surgery, The First Affiliated Hospital of Kunming Medical University, 295 Xichang Rd, Kunming, 650000 China Find articles by Yong Li 1 , Chengkai Cai Chengkai Cai 1 Department of Vascular Surgery, The First Affiliated Hospital of Kunming Medical University, 295 Xichang Rd, Kunming, 650000 China Find articles by Chengkai Cai 1 , Jianyuan Gao Jianyuan Gao 1 Department of Vascular Surgery, The First Affiliated Hospital of Kunming Medical University, 295 Xichang Rd, Kunming, 650000 China Find articles by Jianyuan Gao 1, ✉ Author information Article notes Copyright and License information 1 Department of Vascular Surgery, The First Affiliated Hospital of Kunming Medical University, 295 Xichang Rd, Kunming, 650000 China ✉ Corresponding author. Received 2025 Jul 27; Accepted 2026 Feb 27; Collection date 2026. © The Author(s) 2026 Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ . PMC Copyright notice PMCID: PMC13067403  PMID: 41782121 Abstract Objective To construct a predictive model for post-endovascular restenosis in patients with peripheral artery disease based on a novel strategy that integrates radiomics features with clinical factors, aiming to enhance predictive performance and risk assessment. Methods A retrospective study was conducted on 110 patients with peripheral artery disease who underwent CT angiography and clinical evaluation. The dataset was randomly divided into training and validation sets in a 7:3 ratio. A combined model was developed using multivariable logistic regression. Subsequently, the model was internally validated and a nomogram was constructed. Additionally, models were trained using SVM, Xgboost, KNN, Adaboost, and CatBoost algorithms. The optimal model was assessed through SHAP analysis to evaluate the importance of each feature in predicting restenosis after endovascular treatment for peripheral artery disease. Results After standardization, univariate analysis, and LASSO regression with 5-fold cross-validation for dimensionality reduction, five optimal radiomics features and three clinical factors selected via logistic regression were used to construct the model. Among various models, CatBoost demonstrated the highest predictive performance, with the area under the receiver operating characteristic curve (AUC) for the training and validation sets being 0.985 and 0.878, respectively. SHAP interpretability analysis revealed that the four most important global features influencing the CatBoost model output were, in order, Radiomics Score, PACSS, length, and hypertension. Conclusion The CatBoost model provides superior support for early identification of individuals at higher risk of restenosis after endovascular treatment for peripheral artery disease. It offers a foundation for implementing more targeted prevention and treatment strategies. Keywords: Peripheral artery disease, Machine learning, SHAP, Predictive model, Nomogram Introduction Peripheral arterial disease (PAD) is a major atherosclerotic arterial disorder [ 1 ]. Globally, peripheral artery disease affects approximately 113 million individuals aged 40 years and older, and its prevalence increases with age, having increased by 72% between 1990 and 2019 [ 2 – 6 ]. Lower extremity arterial disease, one of the most common forms of PAD, is characterized by symptoms of limb ischemia, which in severe cases can lead to amputation or death [ 7 ]. Endovascular techniques are widely recognized as minimally invasive procedures that offer precise localization, exact efficacy, and rapid recovery [ 8 ]. Over the past decade, endovascular therapy has become a recommended first-line treatment for symptomatic PAD [ 9 , 10 ]. However, several studies have reported a high incidence of restenosis following endovascular treatment for PAD. Specifically, percutaneous transluminal angioplasty (PTA) has a 3-year patency rate of 48% for mild and 30% for severe occlusive lesions, while stenting achieves 3-year patency rates ranging from 63% to 66% [ 11 ]. Consequently, restenosis has become the most common complication following endovascular therapy for PAD and remains a major clinical challenge and a current focus of therapeutic research [ 12 , 13 ]. Etiology of restenosis following endovascular treatment for PAD is multifactorial and includes hypertension, hyperlipidemia, hyperglycemia, smoking, the inflammatory response, calcification, hemodynamic alterations, and postoperative medication adherence, among other factors (Fig. 1 ) [ 14 – 16 ]. Existing clinical scoring systems fail to comprehensively integrate these diverse risk factors, hindering the ability of clinicians to accurately predict the risk of restenosis after endovascular therapy in patients with PAD. Therefore, developing an accurate prediction model is crucial for the early identification of high-risk individuals, targeted interventions, and informed clinical decision-making. Fig. 1. Open in a new tab Etiology of restenosis following endovascular treatment for PAD Radiomics enables deep analysis and an objective assessment of lesion characteristics by extracting a large number of quantitative features from medical images. It has demonstrated substantial value in the diagnosis, classification, and prognostic evaluation of numerous diseases [ 17 – 20 ], including oncologic, cardiovascular, and neurological conditions [ 21 – 23 ]. Moreover, radiomics allows for the integration of multisource data and modeling complex nonlinear relationships. When combined with artificial intelligence techniques such as machine learning, it serves as a powerful approach for building high-precision predictive models [ 24 – 27 ]. Despite these advancements, current studies predicting postoperative restenosis in PAD mainly depend on the clinical factors to fully exploit the pathophysiological information contained within medical imaging data. Consequently, this study aims to enhance an already existing predictive model for restenosis after endovascular therapy in patients with peripheral arterial disease by integrating radiomics features derived from lower-extremity CTA with key clinical indicators. A CTA can visually depict the location and severity of vascular stenosis while offering detailed hemodynamic information [ 28 , 29 ]. Radiomics features extracted from CTA provide a precise quantitative evaluation of vascular lesions. This proposed model combines radiomics features derived from CTA and key clinical factors for the first time, utilizing a nomogram to visualize individualized risk prediction. Interpretability analysis methods, including SHAP, are employed to reveal the decision-making basis of the model, thus enhancing clinical interpretability and applicability [ 30 – 33 ]. The ultimate goal is to establish a reliable, individualized decision-support tool that enables early detection of high-risk cases of restenosis, supports precision treatment planning, and ultimately improves long-term outcomes in patients with PAD. While previous studies relied primarily on clinical predictors, our study introduces a new integration strategy that incorporates imaging-driven randomics features into an existing clinical framework to improve predictive performance. Materials and methods Study subjects This retrospective study included 110 patients diagnosed with PAD who underwent successful PTA or stent placement between January 2020 and December 2022 at the Department of Vascular Surgery, the First Affiliated Hospital of Kunming Medical University. Patients finally included in the study met the following criteria: (1) diagnosis of lower limb ischemia due to arterial stenosis or occlusion with significant ischemic manifestations such as lower limb claudication and pain (2) Rutherford classification grades 2–6; (3) successful PTA or stenting for the first time; and (4) complete examination data and 2-year survival after surgery and known status of the lower limbs. Exclusion criteria included: (1) any serious health event that could mislead the assessment of lower extremity function, including but not limited to heart failure and lower extremity fractures before admission; (2) lower extremity ischemia due to other etiologies before admission, including but not limited to vasculitis and aneurysms; (3) any previous open surgery or endovascular treatment of the target lower extremity artery; (4) Patients with comorbidities reducing life expectancy to < 1 year or those who died during follow-up from non-disease-related causes (e.g., malignancy, severe cardiovascular or cerebrovascular events); (5) Patients with poor postoperative medication adherence, defined as self-discontinuation, dose reduction, or non-adherence to prescribed medication regimens during follow-up.The Ethics Committee of the First Affiliated Hospital of Kunming Medical University approved this study. Because of the retrospective nature of this study, the Ethics Committee of the First Affiliated Hospital of Kunming Medical University abandoned the written informed consent of all patients. Patient grouping According to the aforementioned inclusion and exclusion criteria, 110 eligible patients were ultimately included in this study. Follow-up continued until the occurrence of restenosis or until 24 months post-treatment, whichever came first. Based on CTA of lower limb arteries and corresponding clinical data, patients were randomly assigned in a 7:3 ratio to a training and validation set. Clinical baseline information and endpoints Medical records were reviewed to collect data on PAD-related risk factors, cardiovascular and cerebrovascular disease history, postoperative medication regimens, demographic information, and blood examination parameters, including red blood cell count, hemoglobin, hematocrit, platelet count, lipid profiles, creatinine, urea, total protein, albumin, globulin, white blood cells, neutrophils, fibrinogen, absolute neutrophil count, absolute lymphocyte count, absolute monocyte count, and platelet-to-lymphocyte ratio. Characteristics of the treated limb included ischemic status and properties of the target lesion, namely lesion type, location, length, calcification severity, intraoperative use of PTA or stents, and residual outflow tract. Calcification was assessed using the Peripheral Arterial Calcium Scoring System (PACSS) [ 34 ] according to the patient’s lower limb CT angiography, and calcification severity was classified as no calcification or mild calcification (grade 0–2) and moderate to severe calcification (grade 3–4) according to the patient’s CT angiography peripheral vascular calcification scoring system.The PACSS score was assessed specifically for the target limb and the corresponding target lesion vessel segment undergoing endovascular therapy. Rutherford classification is divided into intermittent claudication (Rutherford 1–3) and critical limb ischemia (CLI) (Rutherford 4–6) [ 35 ]. The primary endpoint event was restenosis in the follow-up vessels within 24 months after operation. CTA and vascular ultrasonography of lower limb arteries confirmed the test site. The lesion stenosis > 50% was used as the diagnostic criteria for restenosis [ 36 ]. Treatment regimen During endovascular treatment, the approach can be antegrade or retrograde, through the guide wire, catheter through the lesion site, using balloon dilatation or stent implantation for lesion site treatment. Indications for selective stent implantation included residual stenosis > 30% after balloon dilatation, significant or flow-limiting dissection, and thrombosis. An arteriogram was then performed to confirm vessel patency and ensure the absence of thrombus formation, distal embolization, or flow-limiting dissection. Image acquisition CTA imaging of the lower extremity arteries was performed using a Siemens SOMATOM Definition Flash DSCT covering the area from the distal abdominal aorta (approximately L1-L2) to the dorsalis pedis artery. A dose of 80 to 100 mL of the nonionic contrast agent iopromide (360 mg/mL) and 50 mL of normal saline were injected at a rate of 4 to 5 mL/s. The density of the popliteal or iliac region of interest will be detected by the Smart prep method and further detected and evaluated if the density exceeds 100–150 HU. The technical parameters of the scanning instrument include: advanced automatic tube current modulation process, slice thickness 0.625 mm, slice distance 0.625 mm, pitch 0.516 ∶ 1, rotation rate 0.6r/s, 100 kV tube pressure, so as to ensure the accuracy and safety of the scanning instrument. Segmentation of regions of interest and radiomics feature extraction All contrast-enhanced CT images were exported in DICOM format via the Picture Archiving and Communications System. Before feature extraction, preprocessing was performed, including voxel resampling (to 1 × 1 × 1 mm³) and Min-Max normalization. The images were manually delineated ROIs (Fig. 2 ) using the 3D Slicer software package (version 5.6.2) without knowledge of the clinical data by two professional radiologists. The ROI was defined to encompass the entire diseased segment, extending from the proximal adjacent normal vessel segment (with diameter stenosis ≥ 70%) to the distal adjacent normal segment, covering all visible plaques (including calcified, non-calcified, and mixed plaques) and the vessel wall. For patients with multiple lesions or vascular segments, the target lesion was independently selected by two radiologists based on predefined criteria emphasizing imaging features of stenosis (≥ 70%). In cases of disagreement, a consensus was reached through discussion. A complete 3-dimensional ROI was finally generated, and the intraclass correlation coefficient (ICC) was used to evaluate features with good consistency of retention (ICC > 0.80), effectively improving the accuracy and reproducibility of ROI delineation. Fig. 2. Open in a new tab Process of radiomics segmentation. ( A : Original contrast-enhanced CT image. B : Manually delineated ROI highlighting the diseased vessel segment. C : Final 3D reconstruction of the ROI generated using 3D Slicer.) Radiomics module was applied to preprocess images and extract radiomics features. Randomly divide the dataset into the training set and validation set in a ratio of 7:3. In order to improve the accuracy and stability of the model, we optimized the algorithm in terms of data normalization, dimensionality reduction, and feature selection as follows: (1) standardizing the dataset, keeping the data at the same scale and eliminating the impact of radiomics characteristics different magnitudes on the results; (2) in the training set, using univariate analysis of statistically significant judgment based on features with a P value < 0.05; (3) using the least absolute shrinkage and selection operator (LASSO) for the remaining radiomics features, algorithm combined with five-fold cross-validation dimension reduction to select the optimal feature subset, and according to the five selected radiomics features, the radiomics score of each patient was calculated. Model construction Clinical data in the training set were first evaluated using univariate analysis. Subsequently, statistically significant clinical variables underwent multivariate logistic regression analysis to identify independent risk factors for restenosis following endovascular PAD treatment ( P < 0.05) (Table 1 ). Table 1. Clinical factors selected by multivariate regression analysis for restenosis prediction after endovascular PAD treatment Variable B P .value OR PACSS 1.853 0.019 6.380(1.351~30.127) length 2.745 0.014 15.560(1.728~140.076) hypertension 1.757 0.028 5.792(1.204~27.872) Open in a new tab By integrating the radiomics score with clinically independent predictors identified through multivariate logistic regression, a combined predictive model incorporating both radiomics and clinical factors was constructed, and a nomogram was developed for visualization. Based on the clinical factors and enhanced CT image features of the lower limb arteries, five machine learning algorithms, including SVM, Xgboost, KNN, Adaboost, and CatBoost, were used to construct the prediction model. SVM is well known for its ability to handle high‑dimensional data and non‑linear relationships via kernel functions, and it tends to exhibit good generalization performance with limited sample sizes [ 37 ],and helping to avoid issues around overfitting [ 38 ].XGBoost represents a state-of-the-art gradient boosting framework that consistently demonstrates high performance in medical prediction tasks. Its key advantages include regularization to mitigate overfitting, parallel processing capability, and built-in cross-validation mechanisms [ 39 ]. KNN exhibits robustness to noisy data and retains practical value in medical classification, particularly for scenarios with irregular decision boundaries [ 40 ]. AdaBoost has well-documented effectiveness in atherosclerosis-related diagnostic applications, serving as a relevant benchmark for peripheral artery disease research [ 41 ]. CatBoost is characterized by its ability to handle categorical features with minimal preprocessing, offering particular advantages for variables such as hypertension status, PACSS calcification grades, and medication types. It employs ordered target statistics and a novel algorithm for categorical feature processing to reduce target leakage and overfitting [ 42 – 44 ].The accuracies of all the models were compared, and the SHAP method was used to explain the optimal model. Statistical analysis R4.4.2 software and SPSS 27.0 software were used. Measurement data were compared by t-test, and enumeration data were compared by χ 2 test and Fisher exact test. Univariate and multivariate logistic regression analyses were used to screen independent risk factors. Receiver operating characteristic (ROC) curve was plotted. Model power was assessed using Decision Curve Analysis (DCA) with a confusion matrix. P < 0.05 was considered statistically significant. Results Comparison of clinical data In this study, the clinical data of patients in the two groups were selected to compare the basic clinical data of patients after endovascular treatment of PAD in the training set ( n = 76) and the validation set ( n = 34) (as shown in Table 2 ). These patients were further divided into two groups: no restenosis and restenosis. To explore the related factors of restenosis by comparing the basic clinical data, hematological parameters, renal function and lipid parameters of patients. All data were statistically analyzed using the t-test, Chi-square test, or Fisher’s exact test, and P values of less than 0.05 were considered statistically significant. Table 2. Baseline clinical characteristics of patients included in the study Group Training set ( n = 76) Validation set ( n = 34) Variable No restenosis ( N = 63) Restenosis ( N = 13) t/χ² value P value No restenosis( N = 28) Restenosis ( N = 6) t/χ² value P value Age 66.83 ± 12.11 68.69 ± 6.73 -0.537 0.593 65.82 ± 12.80 68.33 ± 6.62 -0.463 0.646 BMI 23.30 ± 6.18 21.59 ± 2.46 0.975 0.333 23.06 ± 3.15 23.55 ± 2.75 -0.352 0.727 RBC 4.48 ± 0.81 4.72 ± 0.84 -0.944 0.348 5.95 ± 7.68 4.18 ± 0.49 0.556 0.582 Hb 134.92 ± 23.96 140.31 ± 25.40 -0.731 0.467 137.29 ± 25.06 134.17 ± 16.99 0.289 0.774 HCT 0.41 ± 0.07 7.11 ± 16.33 -3.346 0.001 0.41 ± 0.07 6.69 ± 15.39 -2.292 0.029 PLT 254.05 ± 143.61 233.15 ± 115.22 0.492 0.624 232.98 ± 81.65 153.00 ± 49.54 2.294 0.029 LDL 2.71 ± 0.91 2.76 ± 0.74 -0.173 0.863 2.64 ± 1.16 2.60 ± 0.74 -0.173 0.863 HDL 0.99 ± 0.24 1.10 ± 0.31 -1.439 0.154 1.02 ± 0.24 1.07 ± 0.30 -0.469 0.642 TG 1.69 ± 0.93 1.87 ± 1.24 -0.62 0.537 1.62 ± 0.72 1.94 ± 0.73 -0.966 0.341 TC 4.57 ± 1.58 4.83 ± 1.36 -0.546 0.587 4.34 ± 1.37 4.36 ± 1.20 -0.032 0.974 Cr 106.35 ± 108.12 88.44 ± 22.69 0.592 0.556 109.68 ± 149.19 93.12 ± 23.85 0.268 0.79 Urea 6.85 ± 3.59 5.92 ± 2.11 0.904 0.369 6.48 ± 3.92 6.03 ± 1.08 0.275 0.785 TP 70.05 ± 5.27 70.17 ± 8.44 -0.069 0.946 72.08 ± 7.74 68.65 ± 11.07 -0.069 0.946 ALB 39.10 ± 5.16 38.17 ± 3.52 0.617 0.539 40.06 ± 3.65 37.20 ± 8.22 1.362 0.183 GLB 31.11 ± 6.01 32.00 ± 9.55 -0.436 0.664 32.02 ± 5.66 31.45 ± 10.69 -0.436 0.664 WBC 8.08 ± 2.93 7.26 ± 2.53 0.939 0.351 7.65 ± 2.72 6.59 ± 2.22 0.939 0.351 NEU 63.87 ± 15.99 63.58 ± 7.26 0.064 0.949 65.45 ± 11.89 71.23 ± 13.44 0.064 0.949 FIB 5.21 ± 9.10 3.21 ± 1.67 0.787 0.434 6.60 ± 12.70 44.43 ± 58.32 -3.255 0.003 ANC 5.49 ± 2.84 4.72 ± 1.94 0.931 0.355 5.49 ± 2.84 4.92 ± 2.29 0.931 0.355 ALC 1.78 ± 0.65 1.64 ± 0.52 0.715 0.477 1.64 ± 0.63 1.11 ± 0.25 1.997 0.054 MONO 0.66 ± 0.83 0.55 ± 0.22 0.481 0.632 0.54 ± 0.22 0.41 ± 0.13 1.365 0.182 PLR 169.86 ± 103.50 157.19 ± 84.36 0.413 0.681 152.14 ± 49.02 141.26 ± 53.33 2.294 0.029 Rutherford 26/37 3/10 1.511 0.348 10/18 1/5 0.819 0.638 PACSS 43/20 3/10 9.206 0.004 20/8 0/6 10.408 0.002 Length 33/30 1/12 8.705 0.004 13/15 1/5 1.807 0.364 Smoking 33/30 6/7 0.167 0.766 11/17 3/3 0.234 0.672 Drink 47/16 11/2 0.598 0.721 21/7 5/1 0.191 1 DP(1) 46/17 11/2 0.773 0.498 24/4 6/0 0.971 0.576 DP(2) 25/38 8/5 2.095 0.219 13/15 5/1 2.701 0.18 Hypertension 36/27 3/10 5.006 0.034 13/15 1/5 1.807 0.364 Diabetes 45/18 9/4 0.025 1 20/8 4/2 0.054 1 Coronaryheartdisease 53/10 9/4 1.591 0.243 26/2 6/0 0.455 1 BTK Arteries 6/6/13/38 0/2/3/8 1.623 0.654 4/1/7/16 0/1/2/3 2.472 0.480 Postoperative medicine 13/25/25 2/1/10 6.619 0.043 8/9/11 1/1/4 1.507 0.615 Equipment 41/22 7/6 0.584 0.532 18/10 3/3 0.427 0.653 Open in a new tab (BMI: Body Mass Index; RBC: Red Blood Cell; Hb: Hemoglobin; HCT: Hematocrit; PLT: Platelet; LDL: Low Density Lipoprotein; HDL: High Density Lipoprotein; TG: Triglyceride; TC: Total Cholesterol; Cr: Creatinine; TP: Total Protein; ALB: Albumin; GLB: Globulin; WBC: White Blood Cell; Neu: Neutrophil; FIB: Fibrinogen; ANC: Absolute Lymphocyte Count; Mono: Monocyte; PLR: Platelet-to-Lymphocyte Ratio; Rutherford: the Rutherford classification was coded as follows: 1 for intermittent claudication (categories 1–3) and 2 for critical limb ischemia (categories 4–6). Length: lesion length; DP(1): the preoperative dorsal pedis artery pulse was defined as follows: 1 for a palpable pulse and 0 for a nonpalpable pulse. DP(2): Postoperative dorsal pedis artery pulse was defined as follows: 1 for a palpable pulse and 0 for a nonpalpable pulse. Postoperative medicine: Postoperative medication: 1 (anticoagulant only), 2(antiplatelet only), and 3 (both). Equipment: Intraoperative strategy: 1 (balloon only), 3 (balloon + stent).) Feature selection and radiomics score establishment Radiomics features included morphological, texture, histogram, and transform-based features [ 45 ]. This study used various methods to process data and select the most predictive features from numerous radiomics parameters. After intraclass correlation coefficient (ICC > 0.8) screening, data normalization, univariate analysis, and dimensionality reduction via least absolute shrinkage and selection operator (LASSO) combined with five-fold cross-validation, five optimal radiomicsfeatures were selected from an initial 1,316 features (Fig. 3 ). The 1,316 radiomics features we extracted included: First-order Statistics༚original_shape_Elongation, original_shape_Flatness, original_shape_LeastAxisLength, etc.; Shape Features༚original_shape_MeshVolume, original_shape_Maximum3DDiameter, etc༛Texture Features༚glcm_Autocorrelation, glcm_Correlation, glcm_ClusterProminence, etc༛Wavelet Transform Features༚wavelet-LLL_glszm_SmallAreaLowGrayLevel Emphasis, wavelet-LLL_ngtdm_Busyness, etc.). These five features included: Fig. 3. Open in a new tab Feature selection using LASSO regression: ( A ) LASSO coefficient profiles showing feature weights distribution(1. “Intercept” is a constant term;2. log.sigma.2.0.mm.3D_glcm_Imc1༛3.Wavelet.HLH_glcm_MCC;4.Wavelet.HHH_glcm_Correlation;5Wavelet. LHH_firstorder_Skewness;6.Wavelet. HLH_firstorder_Kurtosis); ( B ) Optimal λ parameter selection plot; ( C ) Coefficient path demonstrating changes in feature coefficients log.sigma.2.0.mm.3D_glcm_Imc1, wavelet.HLH_glcm_MCC, wavelet.HHH_glcm_Correlation, wavelet.LHH_firstorder_Skewness, wavelet.HLH_firstorder_Kurtosis. Subsequently, a radiomics score was calculated for each patient. Model construction and evaluation We combined three clinically significant factors identified through feature selection with the radiomics score to construct a more accurate predictive model. Multivariate logistic regression analysis was used to construct the prediction model, and ROC curves were plotted for performance evaluation (Fig. 4 ). Results demonstrated an AUC of 0.976 (95% CI: 0.944–1.000) in the training set and 0.833 (95% CI: 0.675–0.992) in the validation set. Decision curve analysis (DCA) was performed to evaluate the net clinical benefit of the predictive model (Fig. 4 ). DCA further confirmed the model’s clinical applicability and predictive value. Fig. 4. Open in a new tab ROC curves and Decision Curve Analysis (DCA) for the multivariate logistic regression predictive model. ( A ) ROC curve of the training set; ( B ) ROC curve of the validation set; ( C ) DCA curve of the training set; ( D ) DCA curve of the validation set To visualize the predicted probability of restenosis within 24 months after endovascular treatment for PAD, a nomogram was constructed based on multivariate regression analysis results (Fig. 5 ). In the nomogram, the length of the scale for each variable reflects the extent of its influence on outcome occurrence, with the radiomics score demonstrating the most significant predictive impact, followed by lesion length, hypertension, and PACSS. Fig. 5. Open in a new tab The nomogram was constructed based on the multivariate logistic regression predictive model incorporating radiomics features and clinical factors In addition, we also construct five prediction models: SVM, Xgboost, KNN, Adaboost, and CatBoost. As shown in Fig. 6 , we evaluate the model performance according to the ROC curve. The training set AUC of SVM, Xgboost, KNN, Adaboost and CatBoost prediction models is 0.957 (95% CI: 0.904-1.000), 0.963(95% CI: 0.963-1.000), 0.982(95% CI: 0.982-1.000), 0.901(95% CI: 0.792-1.000) and 0.985(95% CI: 0.962-1.000), respectively, and the validation set AUC is 0.875(95% CI: 0.717-1.000), 0.848(95% CI: 0.663-1.000), 0.845(95% CI: 0.633-1.000), 0.792 (95% CI: 0.571-1.000)and 0.878(95% CI: 0.681-1.000), respectively. Among these five prediction models, CatBoost model had the highest prediction performance (AUC = 0.878(95% CI: 0.681-1.000)), and the clinical utility and classification performance of the combined model were assessed by plotting the five model decision curves (DCA) (as shown in Fig. 6 ) and confusion matrices (as shown in Figs. 7 and 8 ). According to the results, the training set DCA comparison of the five models showed that the Xgboost, Adaboost, and AdBoost models had higher clinical net benefits. In contrast, the SVM and KNN models had slightly lower clinical net benefits than the other three. The CatBoost, Cataboost, and Xgboost models had higher clinical net benefits in the validation set DCA comparison of the five models. From the confusion matrix of the five models in the training set, it is shown that the true negative examples in SVM, Xgboost, KNN, Adaboost, and CatBoost models are 15.8% and 75%, 14.5% and 81.6%, 17.1% and 77.6%, 14.5% and 76.3%, 15.8% and 80.3%, respectively; the confusion matrix of the five models in the validation set shows that the true negative examples in SVM, Xgboost, KNN, Adaboost, and CatBoost models are 14.7% and 67.6%, 11.8% and 74.9%, 14.7% and 67.6%, 11.8% and 73.5%, 14.7% and 73.5%, respectively, all of which reflect high accuracy. Fig. 6. Open in a new tab ROC and DCA curves for five machine learning prediction models. ( A ) ROC curves for the training set; ( B ) ROC curves for the validation set; ( C ) DCA curves for the training set; ( D ) DCA curves for the validation set Fig. 7. Open in a new tab Confusion matrices for the five machine learning models (training set) Fig. 8. Open in a new tab Confusion matrices for the five machine learning models (validation set) Among the five models, the CatBoost model was selected as optimal based on the highest Area Under the Curve (AUC) in the ROC analysis of the validation set. Using the SHapley Additive exPlanations (SHAP) method, the contributions of each predictor to the CatBoost model for predicting restenosis after endovascular treatment in PAD patients were quantified and ranked in descending order of feature importance (Fig. 9 ). According to SHAP analysis, the radiomics score exhibited the highest mean impact on model prediction outcomes, followed by PACSS, lesion length, and hypertension. Furthermore, SHAP bee-swarm plots were used to visualize positive and negative associations of individual features with outcomes (Fig. 9 ). A feature’s horizontal gradient along the x-axis indicates its positive or negative correlation with prediction outcomes. Color coding denotes the relative magnitude of feature values, with yellow representing higher values and purple representing lower values. The results demonstrated that higher radiomics scores, PACSS, lesion length, and hypertension positively influenced the prediction of restenosis. Fig. 9. Open in a new tab ( A ) SHAP analysis bar plot illustrating the contribution values of each variable in the CatBoost model (ranked by descending importance); ( B ) SHAP analysis bee-swarm plot for visualizing feature contributions in the CatBoost model Discussion PAD of the lower extremities is a common vascular condition that significantly threatens patients’ quality of life and overall health. Endovascular therapy has become the primary treatment modality due to its notable advantages [ 46 , 47 ]. Studies have demonstrated that endovascular treatment can effectively improve lower limb arterial blood supply and reduce the incidence of cerebral infarction, amputation rate, and mortality in PAD patients compared with drug treatment [ 48 ]. Lower extremity arterial CTA is of great significance in diagnosing PAD restenosis. This imaging method is of high value in assessing the blood flow of lower extremity arteries, the location and severity of stenosis, and other lesion information [ 17 , 18 , 49 ], but the traditional visual assessment method is not only subjectively limited, but also challenging to excavate and analyze the quantitative characteristics of its images. Radiomics improves the predictive performance of medical images by extracting quantitative features of lesions through computers [ 31 ]. Numerous studies have shown that the application and value of radiomics in various malignant tumors such as lung cancer, breast cancer, and liver cancer have been confirmed [ 28 , 29 , 32 ]. Thus, we believe that radiomics features extracted from lower extremity arterial CTA images could predict restenosis after endovascular treatment for PAD. Machine learning, a subset of artificial intelligence, fits predictive models to data or classifies data through information analysis [ 33 ]. Previous studies have relied solely on clinical factors to predict restenosis after endovascular treatment for PAD. However, clinical practice reveals that imaging characteristics, particularly findings from lower extremity arterial CTA, also hold critical prognostic value. Therefore, our study is the first to integrate clinical parameters with radiomics features to construct a combined prediction model, employing SHAP to interpret the optimal CatBoost model, significantly enhancing predictive performance for restenosis following PAD intervention. Numerous studies have highlighted the association of hypertension, arterial calcification severity, and lesion length with restenosis after PAD endovascular intervention [ 50 – 52 ], aligning closely with our findings. Multivariate logistic regression was used to build the combined model in our study. A nomogram combining the radiomics score and clinical parameters was used to represent each variable in the multivariate regression model by setting a ruler score. Finally, a total score was calculated to predict patients’ restenosis probability. PACSS score, lesion length, hypertension and radiomics score were key variables in this predictive model. Its importance ranking is once: radiomics score, length, hypertension, PACSS. The primary strength of the nomogram lies in its intuitive and convenient nature, enabling clinicians to assess the risk of restenosis based on patient-specific characteristics rapidly. For instance, physicians can estimate restenosis risk by evaluating a patient’s PACSS score, imaging features, and hypertensive status. Such predictive tools facilitate personalized treatment decisions for high-risk patients, determining the need for closer follow-up, modifying treatment regimens, or selecting more suitable therapeutic approaches. In this retrospective study, we also developed five machine learning algorithms: SVM, Xgboost, KNN, Adaboost, and CatBoost to predict the occurrence of restenosis after endovascular treatment of PAD. CatBoost is an algorithm based on a gradient boosting tree (GBDT), which has advantages for the processing of classification and heterogeneous data, avoiding overfitting, automatic parameter adjustment, and capturing nonlinear relationships in data, especially for complex, high-dimensional, or noisy data [ 42 – 44 ]. In our study, the CatBoost model performed better than the SVM, Xgboost, KNN, and Adaboost models and performed internal validation. The model was evaluated with area under the receiver operating characteristic curve (AUC), decision curve analysis (DCA), and confusion matrix. As a well-established method for evaluating the clinical utility of predictive models, DCA quantifies the net benefit of using a model across different threshold probabilities. The results of this study demonstrated that the CatBoost model achieved a positive net benefit in both the training and validation sets (Fig. 6 C and D). This suggests that implementing this model for risk stratification in clinical practice may help optimize treatment decisions—for example, by intensifying follow-up for high-risk patients or selecting more aggressive interventional strategies—while avoiding overtreatment in low-risk patients.By handling categorical variables (such as PACSS grade and hypertension) through ordered target statistics, the CatBoost model avoids the information loss associated with one‑hot encoding and preserves more predictive signals. Moreover, CatBoost’s symmetric tree structure and ordered boosting process provide built‑in regularization mechanisms that may help mitigate overfitting. Empirical results demonstrate that CatBoost achieved a validation set AUC of 0.878, with a relatively small AUC difference between the training and validation sets (Δ = 0.107), indicating good generalization ability. Although the SVM model exhibited a slightly smaller drop (Δ = 0.082), its validation AUC was lower (0.875), suggesting relatively weaker overall discriminative power. The confusion matrices further indicate that CatBoost achieved a more balanced sensitivity and specificity in the validation set, reflecting more stable predictive performance. Decision curve analysis (DCA) revealed that both CatBoost and Xgboost exhibited favorable clinical utility, effectively identifying high‑risk patients and aiding clinical decision‑making. Model predictions were interpreted using the SHAP method to eliminate the black box effect of machine learning and visualize the contribution of each feature to the prediction outcome, helping clinicians to understand better the decision-making process of the model [ 53 – 55 ]. An interpretable analysis of the optimal model CatBoost can be seen in Fig. 9 , consistent with previous studies and clinical experience that three clinical variables hypertension, lesion length, and calcification degree in this model are essential in predicting patients with restenosis after endovascular PAD treatment, and the radiomics characteristics included in our study have the most significant contribution to the outcome of this model as core predictors of the model, suggesting that our imaging characteristics (e.g., stenosis degree, calcification distribution) can indirectly reflect the vascular biological status and help identify the clinical significance of lesions prone to restenosis. This chart quantifies the predictive contribution of each feature in the CatBoost model by SHAP values to provide personalized treatment recommendations for patients, with radiomics scores and lesion lengths as core drivers, consistent with high-risk factors for the situation in clinical practice, and requires attention when it is higher radiomics score and longer lesion length, while taking measures to control the further development of calcified vessels and control the stability of blood pressure will have a positive impact on patients after endovascular PAD treatment. In the future, this model can be used to optimize the preoperative evaluation process and realize the dynamic monitoring and precise intervention of restenosis risk. This study not only fills the gap in the application of Radiomics in the field of PAD, but also provides novel insights into the risk prediction of restenosis after endovascular treatment.Specifically, the CatBoost model, constructed based on a novel strategy that integrates radiomics features and clinical factors, facilitates more targeted clinical practice, including preoperative assessment, postoperative management, and physician-patient communication. In terms of preoperative assessment, by integrating the radiomics score, PACSS calcification grade, lesion length, and hypertension status, clinicians can rapidly identify patients at high risk of restenosis and consider the use of targeted therapeutic devices during preoperative discussions—such as atherectomy, rotational ablation, or laser ablation—depending on plaque characteristics. For patients with less severe plaque burden, balloon angioplasty or stent implantation may be considered, with the selective use of drug-coated balloons based on individual circumstances. Regarding postoperative management, based on the predicted risk from the model, individualized follow-up intervals can be established for patients—for example, high-risk patients may undergo CTA every 6 months after the first year—along with tailored secondary prevention strategies (e.g., strict blood pressure control).And through the automated machine learning model, clinicians do not need to analyze many image data and medical record information manually, and can quickly obtain reliable prediction results through the model. This saves a lot of time and resources and improves the efficiency of diagnosis and treatment. The SHAP analysis can provide interpretability of model predictions and guide physicians to understand the specific effects of various clinical factors on the occurrence of restenosis. This increases trust in the model and guides physicians to communicate with patients and explain the basis for treatment decisions. Models combining machine learning algorithms and radiomics features in future medical care have significant generalization value. With the popularization of image data and the continuous development of machine learning technology, this model can gradually realize automatic analysis through integration with electronic medical record systems and image analysis platforms. However, this study was unable to incorporate dynamic variables such as new-onset clinical events (e.g., myocardial infarction) or medication adjustments into the predictive model, as these time-dependent covariates may influence the occurrence of restenosis. Future prospective studies should collect such dynamic information, and this work highlights the need for prospective, dynamic risk prediction studies. With the advancement of multi-center prospective research, the accuracy and generalizability of the model are expected to further improve. Ultimately, it holds promise for widespread application in clinical practice across different hospitals and regions, promoting the popularization and development of personalized treatment for patients with PAD. Limitations Our study has several limitations. First, it was a single-center retrospective study, and the limited number of restenosis cases in the validation set may render performance metrics such as the AUC susceptible to bias.In our study, we constructed a machine learning model that, for the first time, integrates radiomics features and clinical factors to predict restenosis following endovascular treatment in PAD patients, achieving encouraging predictive results despite the relatively small sample size. However, further studies with larger sample sizes, prospective designs, and multicenter validation are necessary to optimize the performance of the model. Additionally, our predictive model lacks external validation; thus, further external datasets are required to verify its generalizability and clinical applicability. Conclusion In this study, we successfully constructed the CatBoost model, an optimal prediction model integrating radiomics and clinical factors to predict restenosis following PAD endovascular treatment. Furthermore, interpretable machine learning methods can identify risk factors influencing restenosis after endovascular treatment for PAD, thereby assisting clinicians in developing personalized strategies for treatment, prevention, and long-term prognosis management. Acknowledgements We thank Kunming Medical University for the support in this manuscript preparation–grant number 2024S076. Abbreviations AUC Area under the curve CLI Critical limb ischemia CT Computed tomography CTA Computed Tomography Angiography DCA Decision Curve Analysis KNN K-Nearest Neighbors LASSO Least Absolute Shrinkage and Selection Operator MALE Major adverse limb events MR Magnetic resonance PACS Picture Archiving and Communication System PAD Peripheral arteries disease PTA Percutaneous Transluminal Angioplasty ROI Region of Interest ROC Receiver Operating Characteristic Curve SHAP SHapley Additive exPlanations SVM Support Vector Machine Author contributions Jianyuan Gao: Writing – review & editing, Funding acquisition; Min Luo: Writing original draft, visualization, validation, supervision; Haohua Wang: Project administration, methodology; Hai Xia: Software, resources; Wangxing Feng: Investigation; Bin Li, Yong Li, and Chenkai Cai: Formal analysis, data curation, conceptualization. Funding This project was funded by the Kunming Medical University Project (No. 2024S076). Data availability The data on which the study is based were accessed from a repository and areavailable for downloading through the following link.（ https://dataverse.harvard.edu/dataset.xhtml?persistent ). Declarations Ethics approval and consent to participate The study was conducted following the Declaration of Helsinki and was approved by the Ethics Committee of The First Affiliated Hospital of Kunming Medical University, and the requirement for written informed consent was waived. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References 1. Kullo IJ, Rooke TW. CLINICAL PRACTICE. Peripheral Artery Disease [J]. N Engl J Med. 2016;374(9):861–71. [ DOI ] [ PubMed ] [ Google Scholar ] 2. Global burden. of 369 diseases and injuries in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019 [J]. Lancet. 2020;396(10258):1204–22. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 3. Lin J, Chen Y, Jiang N, Li Z, Xu S. Burden of Peripheral Artery Disease and Its Attributable Risk Factors in 204 Countries and Territories From 1990 to 2019 [J]. Front Cardiovasc Med. 2022;9:868370. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 4. Global burden of peripheral artery disease and its risk factors. 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019 [J]. Lancet Glob Health. 2023;11(10):e1553–65. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Liu W, Yang C, Chen Z, Lei F, Qin J J J, Liu H, et al. Global death burden and attributable risk factors of peripheral artery disease by age, sex, SDI regions, and countries from 1990 to 2030: Results from the Global Burden of Disease study 2019. Atherosclerosis. 2022;347:17–27. [ DOI ] [ PubMed ] [ Google Scholar ] 6. Roth GA, Mensah GA, Johnson CO, Addolorato G, Ammirati E, Baddour LM, et al. Global Burden of Cardiovascular Diseases and Risk Factors, 1990–2019: Update From the GBD 2019 Study [J]. J Am Coll Cardiol. 2020;76(25):2982–3021. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 7. Ye M, Qian X, Guo X, Wang H, Ni Q, Zhao Y, et al. Neutrophil-Lymphocyte Ratio and Platelet-Lymphocyte Ratio Predict Severity and Prognosis of Lower Limb Arteriosclerosis Obliterans [J]. Ann Vasc Surg. 2020;64:221–7. [ DOI ] [ PubMed ] [ Google Scholar ] 8. Yao W, Wang L, Chen Q, Wang F, Feng N. Effects of Valsartan on Restenosis in Patients with Arteriosclerosis Obliterans of the Lower Extremities Undergoing Interventional Therapy: A Prospective, Randomized, Single-Blind Trial [J]. Med Sci Monit. 2020;26:e919977. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 9. Aboyans V, Ricco JB, Bartelink MEL, Björck M, Brodmann M, Cohnert T, et al. 2017 ESC Guidelines on the Diagnosis and Treatment of Peripheral Arterial Diseases, in collaboration with the European Society for Vascular Surgery (ESVS): Document covering atherosclerotic disease of extracranial carotid and vertebral, mesenteric, renal, upper and lower extremity arteriesEndorsed by: the European Stroke Organization (ESO)The Task Force for the Diagnosis and Treatment of Peripheral Arterial Diseases of the European Society of Cardiology (ESC) and of the European Society for Vascular Surgery (ESVS) [J]. Eur Heart J. 2018;39(9):763–816. [ DOI ] [ PubMed ] [ Google Scholar ] 10. Gerhard-Herman MD, Gornik HL, Barrett C, Barshes NR, Corriere MA, Drachman DE, et al. 2016 AHA/ACC Guideline on the Management of Patients With Lower Extremity Peripheral Artery Disease: A Report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines [J]. J Am Coll Cardiol. 2017;69(11):e71–126. [ DOI ] [ PubMed ] [ Google Scholar ] 11. Scott EC, Biuckians A, Light RE, Burgess J, Meier GH. Subintimal angioplasty: Our experience in the treatment of 506 infrainguinal arterial occlusions [J]. J Vasc Surg. 2008;48(4):878–84. 3rd,et al. [ DOI ] [ PubMed ] [ Google Scholar ] 12. Liistro F, Reccia MR, Angioli P, Ducci K, Ventoruzzo G, Falsini G, Scatena A, Pieroni M, Bolognese L. Drug-Eluting Balloon for Below the Knee Angioplasty: Five-Year Outcome of the DEBATE-BTK Randomized Clinical Trial [J]. Cardiovasc Intervent Radiol. 2022;45(6):761–9. [ DOI ] [ PubMed ] [ Google Scholar ] 13. Almasri J, Adusumalli J, Asi N, Lakis S, Alsawas M, Prokop LJ, et al. A systematic review and meta-analysis of revascularization outcomes of infrainguinal chronic limb-threatening ischemia [J]. J Vasc Surg. 2019;69(6s):s126–36. [ DOI ] [ PubMed ] [ Google Scholar ] 14. Shah PK. Inflammation, infection and atherosclerosis [J]. Trends Cardiovasc Med. 2019;29(8):468–72. [ DOI ] [ PubMed ] [ Google Scholar ] 15. Eraso LH, Fukaya E, Mohler ER 3rd, Xie D, Sha D, et al. Peripheral arterial disease, prevalence and cumulative risk factor profile analysis [J]. Eur J Prev Cardiol. 2014;21(6):704–11. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Ong SB, Hernández-Reséndiz S, Crespo-Avilan GE, Mukhametshina RT, Kwek XY, Cabrera-Fuentes HA, Hausenloy DJ. Inflammation following acute myocardial infarction: Multiple players, dynamic roles, and novel therapeutic opportunities [J]. Pharmacol Ther. 2018;186:73–87. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 17. Van Cester TmerenJE, Tanadini-Lang D, Alkadhi S, Baessler H. Radiomics in medical imaging-how-to guide and critical reflection [J]. Insights Imaging. 2020;11(1):91. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 18. Rogers W, Thulasi Seetha S, Refaee T, a G, Lieverse RIY, Granzier RWY, Ibrahim A, et al. Radiomics: from qualitative to quantitative imaging [J]. Br J Radiol. 2020;93(1108):20190948. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 19. Yang Y, Tang L, Deng Y, Li X, Luo A, Zhang Z, et al. The predictive performance of artificial intelligence on the outcome of stroke: a systematic review and meta-analysis [J]. Front Neurosci. 2023;17:1256592. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 20. Chen Q, Xia T, Zhang M, Xia N, Liu J, Yang Y. Radiomics in Stroke Neuroimaging: Techniques, Applications, and Challenges [J]. Aging Dis. 2021;12(1):143–54. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 21. Xing X, Li L, Sun M, Yang J, Zhu X, Peng F, et al. Deep-learning-based 3D super-resolution CT radiomics model: Predict the possibility of the micropapillary/solid component of lung adenocarcinoma [J]. Heliyon. 2024;10(13):e34163. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Talebi A, Bitarafan-Rajabi A, Alizadeh-Asl A, Seilani P, Khajetash B, Hajianfar G, et al. Machine learning based radiomics model to predict radiotherapy induced cardiotoxicity in breast cancer [J]. J Appl Clin Med Phys. 2024: e14614. [ DOI ] [ PMC free article ] [ PubMed ] 23. Guo K, Zhu B, Li R, Xi J, Wang Q, Chen K, et al. Machine learning-based nomogram: integrating MRI radiomics and clinical indicators for prognostic assessment in acute ischemic stroke [J]. Front Neurol. 2024;15:1379031. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 24. Gui CP, Chen YH, Zhao HW, Cao JZ, Liu TJ, Xiong SW, et al. Multimodal recurrence scoring system for prediction of clear cell renal cell carcinoma outcome: a discovery and validation study [J]. Lancet Digit Health. 2023;5(8):e515–24. [ DOI ] [ PubMed ] [ Google Scholar ] 25. Chen RJ, Lu MY, Williamson DFK, Chen TY, Lipkova J, Noor Z, et al. Pan-cancer integrative histology-genomic analysis via multimodal deep learning [J]. Cancer Cell. 2022;40(8): 865 – 78.e6. [ DOI ] [ PMC free article ] [ PubMed ] 26. Boehm KM, Aherne EA, Ellenson L, Nikolovski I, Alghamdi M, Vázquez-García I, et al. Multimodal data integration using machine learning improves risk stratification of high-grade serous ovarian cancer [J]. Nat Cancer. 2022;3(6):723–33. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 27. Sammut SJ, Crispin-Ortuzar M, Chin SF, Provenzano E, Bardwell HA, Ma W, et al. Multi-omic machine learning predictor of breast cancer therapy response [J]. Nature. 2022;601(7894):623–9. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 28. Fan ZY, Wang XY. Diagnostic approach and structured reporting of CTA in lower extremity arteriosclerosis obliterans[J]. Radiologic Pract. 2017;32(12): 1300–5. 29. Guo L, Lin ZY, Yang M, Liu JX, Wang XY. Feasibility study of 70kVp combined with individualized contrast agent injection protocol in lower extremity CTA for diabetic foot[J]. Radiologic Pract. 2016;31(02):118–22. [ Google Scholar ] 30. Balachandran VP, Gonen M, Smith JJ, Dematteo RP. Nomograms in oncology: more than meets the eye [J]. Lancet Oncol. 2015;16(4):e173–80. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 31. Yao W, Yang S, Ge Y, Fan W, Xiang L, Wan Y, et al. Computed Tomography Radiomics-Based Prediction of Microvascular Invasion in Hepatocellular Carcinoma [J]. Front Med (Lausanne). 2022;9:819670. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 32. Fan R, Long X, Chen X, Wang Y, Chen D, Zhou R. The Value of machine learning-based radiomics model characterized by pet imaging with (68)Ga-FAPI in assessing microvascular invasion of hepatocellular carcinoma [J]. Acad Radiol. 2024. [ DOI ] [ PubMed ] 33. Greener JG, Kandathil SM, Moffat L, Jones DT. A guide to machine learning for biologists [J]. Nat Rev Mol Cell Biol. 2022;23(1):40–55. [ DOI ] [ PubMed ] [ Google Scholar ] 34. Rocha-Singh KJ, Zeller T, Jaff MR. Peripheral arterial calcification: prevalence, mechanism, detection, and clinical implications [J]. Catheter Cardiovasc Interv. 2014;83(6):E212–20. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 35. Tao PAN, Shiyun TIAN, Tao ZHANG. JI Donghua. The effect of anemia on restenosis in patients with femoropopliteal atherosclerotic occlusion after drug-coated balloon angioplasty[J]. J Intervent Radiol. 2021;30(08): 799–803. 36. Li Z, Weibin W, Mian W, Zhengde Z, Yunyan L, Chen Y, et al. Effect of directional atherectomy system in treatment of femoropopliteal artery in-stent restenosis[J]. J Vascular Endovascular Surg. 2022;8(03):263–7. [ Google Scholar ] 37. Hu X, Wong KK, Young GS, Guo L, Wong ST. Support vector machine multiparametric MRI identification of pseudoprogression from tumor recurrence in patients with resected glioblastoma [J]. J Magn Reson Imaging. 2011;33(2):296–305. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 38. Guido R, Ferrisi S, Lofaro D, Conforti D. An Overview on the Advancements of Support Vector Machine Models in Healthcare Applications: A Review[J]. Information. 2024;15(4):235. [ Google Scholar ] 39. Yi F, Yang H, Chen D, Qin Y, Han H, Cui J, Bai W, Ma Y, Zhang R, Yu H. XGBoost-SHAP-based interpretable diagnostic framework for alzheimer’s disease [J]. BMC Med Inf Decis Mak. 2023;23(1):137. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 40. Ehsani R, Drabløs F. Robust Distance Measures for kNN Classification of Cancer Data [J]. Cancer Inf. 2020;19:1176935120965542. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 41. Liu L, Bi B, Cao L, Gui M, Ju F. Predictive model and risk analysis for peripheral vascular disease in type 2 diabetes mellitus patients using machine learning and shapley additive explanation [J]. Front Endocrinol (Lausanne). 2024;15:1320335. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 42. Prokhorenkova L, Gusev G, Vorobev A, Dorogush AV, Gulin A. CatBoost: unbiased boosting with categorical features [J]. 2017. 43. Hancock JT, Khoshgoftaar TM. CatBoost for big data: an interdisciplinary review [J]. J Big Data. 2020;7(1):94. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 44. Wang X, Yang L, Wang R, mRCat:. A Novel catboost predictor for the binary classification of mRNA subcellular localization by fusing large language model representation and sequence features [J]. Biomolecules. 2024;14(7). [ DOI ] [ PMC free article ] [ PubMed ] 45. Mayerhoefer ME, Materka A, Langs G, Häggström I, Szczypiński P, Gibbs P, Cook G. Introduction to Radiomics [J]. J Nucl Med. 2020;61(4):488–95. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 46. Zhou T, Sun. Z C.Efficacy of chocolate balloon angioplasty in treating lower extremity arteriosclerosis obliterans in elderly patients and its effects on arteriosclerosis indicators and hemodynamicsp[J]. Chin J Gerontol. 2024;44(01):72–5. [ Google Scholar ] 47. Shuangshuang LU, Zun LIU, Weihao SHI, Lei ZHU, Bing LENG, Weijian FAN, Xiaosheng CUI, et al. Exercise and Collateral Vessels of Patients with Arteriosclerosis Obliterans after Endovascular Treatment: A Cross⁃sectional Study[J]. Chin Comput Med Imag. 2023;29(02):191–5. [ Google Scholar ] 48. Li Y. Zhao Zhigang,He Hongbo,Ni Yinxing,Sun Fang,Zhong Jian,et al.Long-term role of endovascular intervention in patients with peripheral arterial disease and its influencing factors[J]. J Third Mil Med Univ. 2013;35(08):789–92. [ Google Scholar ] 49. Meyersohn NM, Walker TG, Oliveira GR. Advances in axial imaging of peripheral vascular disease [J]. Curr Cardiol Rep. 2015;17(10):87. [ DOI ] [ PubMed ] [ Google Scholar ] 50. Wen-tao Tang J, Liu. Zhang-lun Xu.Effect of interventional therapy and conservative therapy on arteriosclerosis obliterans of lower limbs and prognostic risk factors analysis. China J Mod Med. 2022;32(05): 93–100. 51. Tepe G, Beschorner U, Ruether C, Fischer I, Pfaffinger P, Noory E, et al. Drug-Eluting Balloon Therapy for Femoropopliteal Occlusive Disease: Predictors of Outcome With a Special Emphasis on Calcium [J]. J Endovasc Ther. 2015;22(5):727–33. [ DOI ] [ PubMed ] [ Google Scholar ] 52. Yuchi Zou, Qiang Tong, Xuehu Wang, Xinyi Li, Yu Zhao. Cheng Jun. Development and validation of a nomogram for predicting restenosis in 12 months after percutaneous transluminal angioplasty of arteriosclerosis obliterans[J]. J Army Med Univ. 2023;45(07): 705–14. 53. Lundberg S, Lee SI. A unified approach to interpreting model predictions [J]. 2017. 54. Lundberg SM, Erion G, Chen H, Degrave A, Prutkin JM, Nair B, et al. From Local Explanations to Global Understanding with Explainable AI for Trees [J]. Nat Mach Intell. 2020;2(1):56–67. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 55. Nohara Y, Matsumoto K, Soejima H, Nakashima N. Explanation of machine learning models using shapley additive explanation and application for real data in hospital [J]. Comput Methods Programs Biomed. 2022;214:106584. [ DOI ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Data Availability Statement The data on which the study is based were accessed from a repository and areavailable for downloading through the following link.（ https://dataverse.harvard.edu/dataset.xhtml?persistent ). Articles from BMC Medical Informatics and Decision Making are provided here courtesy of BMC ACTIONS View on publisher site PDF (3.2 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