Revolutionizing Heart Failure Management With Artificial Intelligence: A Narrative Review of Diagnostic, Prognostic, and Therapeutic Innovations - 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 Health Sci Rep . 2026 Mar 30;9(4):e71855. doi: 10.1002/hsr2.71855 Search in PMC Search in PubMed View in NLM Catalog Add to search Revolutionizing Heart Failure Management With Artificial Intelligence: A Narrative Review of Diagnostic, Prognostic, and Therapeutic Innovations Farrukh Ansar Farrukh Ansar 1 Alkhidmat Raazi Hospital, Islamabad, Pakistan Find articles by Farrukh Ansar 1 , Muhammad Aamir Waheed Muhammad Aamir Waheed 2 Hamad General Hospital/Qatar University, Doha, Qatar Find articles by Muhammad Aamir Waheed 2, ✉ , Usman Zafar Usman Zafar 1 Alkhidmat Raazi Hospital, Islamabad, Pakistan Find articles by Usman Zafar 1 , Abdulrahman Kolapo Abdulrahman Kolapo 3 Lincoln County Hospital, Lincoln, UK Find articles by Abdulrahman Kolapo 3 , Walid Sarfaraz Walid Sarfaraz 4 North Cumbria Integrated Care NHS Trust, UK Find articles by Walid Sarfaraz 4 , Khalid Rashid Khalid Rashid 5 University Hospital North Tees and Hartpool NHS Trust, UK Find articles by Khalid Rashid 5 Author information Article notes Copyright and License information 1 Alkhidmat Raazi Hospital, Islamabad, Pakistan 2 Hamad General Hospital/Qatar University, Doha, Qatar 3 Lincoln County Hospital, Lincoln, UK 4 North Cumbria Integrated Care NHS Trust, UK 5 University Hospital North Tees and Hartpool NHS Trust, UK * Correspondence: Muhammad Aamir Waheed ( [email protected] ) ✉ Corresponding author. Revised 2025 Dec 10; Received 2025 Jul 8; Accepted 2026 Feb 11; Collection date 2026 Apr. © 2026 The Author(s). Health Science Reports published by Wiley Periodicals LLC. This is an open access article under the terms of the http://creativecommons.org/licenses/by-nc-nd/4.0/ License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non‐commercial and no modifications or adaptations are made. PMC Copyright notice PMCID: PMC13087617 PMID: 42005648 ABSTRACT Background and Aims Heart failure (HF) remains a major global health burden, affecting over 64 million individuals worldwide. Early detection and optimal management are often limited by subjective interpretation of diagnostic tests and variable clinical expertise. Artificial intelligence (AI) has emerged as a transformative technology that can enhance diagnostic precision, risk stratification, and therapeutic decision‐making. This review aims to synthesize current evidence on clinically validated AI applications across the HF care continuum. Methods A narrative review of recent peer‐reviewed studies was conducted to evaluate AI‐based tools applied in electrocardiography (ECG), echocardiography, cardiac magnetic resonance (CMR), remote monitoring, and smart devices. Emphasis was placed on studies reporting validated performance metrics such as area under the curve (AUC), sensitivity, and specificity, and those integrated into clinical workflows or approved by regulatory bodies. Results AI‐enhanced ECG models have demonstrated high diagnostic accuracy for left‐ventricular systolic dysfunction and diastolic impairment, with AUC values up to 0.92; surpassing traditional risk scores. In cardiac imaging, deep‐learning systems now automate ejection‐fraction and diastolic‐function quantification with precision comparable to expert readers. AI‐driven platforms such as EchoGo and PanEcho enable efficient and consistent image interpretation, while wearables and implantable sensors like HeartLogic and CardioMEMS provide real‐time hemodynamic monitoring and predict decompensation several days before clinical deterioration (sensitivity 70%–88%). Over 40 AI‐based cardiovascular tools have received regulatory clearance, supporting their translational maturity. Conclusion AI technologies are redefining HF care by enabling earlier diagnosis, individualized therapy, and proactive monitoring. However, challenges persist regarding data diversity, model transparency, and clinical integration. 1. Introduction Heart failure (HF) is a major global health crisis affecting about 64 million people worldwide [ 1 ]. It remains a leading cause of hospitalization, morbidity, and mortality, imposing a heavy burden on healthcare systems [ 1 ]. HF is a complex syndrome marked by impaired cardiac pumping function and presents in multiple forms. Roughly half of all HF patients have preserved ejection fraction (HFpEF) [ 2 ]. HFpEF is especially challenging to diagnose, often underrecognized due to nonspecific symptoms and the multifactorial criteria required for confirmation [ 2 ]. Common symptoms such as shortness of breath, fatigue, and edema, overlap with many other conditions, complicating early identification [ 2 ]. Diagnosis relies on combined clinical features, biomarkers, and imaging findings, which may not consistently yield clear results [ 3 ]. Traditional diagnostic methods including echocardiography, biomarker assays, and clinical scoring systems remain essential in HF management but have notable limitations [ 4 ]. Echocardiography may miss subtle functional abnormalities, while biomarker assays are often unavailable in resource‐limited settings [ 5 ]. Clinical scoring systems rely heavily on physician judgment, influenced by individual experience [ 6 ]. These challenges highlight the need for more efficient, accurate, and reliable diagnostic tools for HF. Recent advances in artificial intelligence (AI), particularly deep learning, offer promising solutions for HF management [ 7 ]. In HF care, AI shows strong potential to enhance diagnosis, prognostication, and therapy optimization [ 8 ]. AI systems can refine risk stratification, predicting hospital readmission, decompensation, and mortality [ 8 ]. They also support personalized treatment by enabling real‐time therapy adjustments based on patient data [ 7 ]. Furthermore, AI‐driven mobile health, remote monitoring, and telehealth platforms enhance continuity of care and patient engagement [ 9 ]. Comparable innovations are transforming other cardiology fields, including coronary artery imaging, arrhythmia detection, and electrophysiologic mapping [ 9 ]. This review explores the role of AI in contemporary HF management, emphasizing clinically relevant applications. Key domains include ECG interpretation, cardiovascular imaging, remote monitoring, mobile health, telehealth, and smart medical devices. We highlight recent studies and AI systems already deployed or nearing integration into clinical workflows, focusing on architectures such as convolutional neural networks (CNNs) and transformers, and their performance metrics, including AUROC, sensitivity, and specificity. By linking technical innovation with clinical applicability, the review offers actionable insights for clinicians navigating the evolving AI landscape in HF care. Figure 1 provides an overview of how AI contributes to heart‐failure detection, monitoring, and management across the continuum of care. Figure 1. Open in a new tab Overview of artificial‐intelligence applications across the heart‐failure care pathway. 2. AI in ECG Interpretation: Unlocking the Potential for Early Detection of Heart Failure The 12‐lead electrocardiogram (ECG) is one of the most widely used diagnostic tools in clinical cardiology due to its low cost, non‐invasive nature, and ability to capture a wealth of information about the electrical activity of the heart. It is especially useful for identifying arrhythmias, ischemia, and other cardiac conditions [ 10 ]. However, subtle markers of cardiac dysfunction, particularly those indicative of HF, often remain undetected using traditional methods of interpreting ECGs [ 11 ]. AI, particularly deep learning, has the potential to uncover hidden patterns in ECGs, allowing for earlier detection of cardiac dysfunction, sometimes even before symptoms appear [ 12 ]. One of the most compelling uses of AI in ECG interpretation is the detection of left‐ventricular ejection fraction (LVEF), a key indicator of systolic heart failure. Attia et al. developed a CNN model to identify reduced LVEF from standard 12‐lead ECGs [ 13 ]. In a prospective Mayo Clinic validation including 3,874 patients with paired ECG and transthoracic echocardiography (TTE), the model achieved 82.5% sensitivity, 86.8% specificity, and an AUC of 0.918 for LVEF ≤ 35%. The so‐called “AI‐ECG” could thus flag asymptomatic individuals for confirmatory imaging. Notably, many false‐positive cases had borderline LVEF (36%–40%), suggesting detection of subclinical dysfunction. Although a high AUC reflects strong discrimination, it does not equate to outcome improvement; external multi‐ethnic validation remains essential. These deep architectures capture complex temporal–spatial ECG patterns beyond conventional analytics but require rigorous control of overfitting. Evidence from primary‐care and low‐resource settings is still limited, so adoption should remain context‐specific pending large pragmatic trials. AI's application in ECG extends beyond detecting systolic dysfunction. Lee et al. developed a deep neural network capable of estimating diastolic function parameters, including left ventricular filling pressures, directly from ECG data [ 14 ]. The model achieved an AUC of 0.91 (95% CI: 0.909–0.914), with 83.2% sensitivity and 82.9% specificity, and accurately graded diastolic dysfunction (grades 1–3). These findings underscore AI‐ECG's ability to capture subtle electrophysiologic features of diastolic heart failure, a hallmark of HFpEF, and its potential as an adjunctive or triage tool to identify patients needing echocardiographic assessment. Beyond diagnosis, AI‐ECG systems can also predict future HF risk. In a large multinational study, an AI model trained to detect LV systolic dysfunction was applied to healthy individuals and followed over 3–4 years [ 15 ]. Those flagged “at risk” had a 4–24‐fold higher incidence of HF‐related hospitalizations, with hazard ratios from 3.9 (US) to 23.5 (Brazil). When combined with clinical risk scores, the model achieved C‐statistics of 0.718–0.810 [ 15 ]. These results suggest AI‐processed ECGs may function as “digital biomarkers,” enabling early identification and preventive care for individuals at risk of heart failure. AI‐assisted ECG interpretation represents a pivotal evolution in cardiovascular care. By uncovering previously inaccessible patterns, AI can enhance diagnosis, predict risk, and guide therapy in heart failure. As these tools mature, they are poised to transform HF detection through more proactive and personalized care. Maximizing their impact, however, requires standardized systems for ECG acquisition, storage, and interpretation across healthcare settings. Variability in hardware, software formats, electronic health records (EHRs), and data labeling limits interoperability and degrades algorithm performance. A unified framework for ECG data management would improve reproducibility, support large‐scale model training, and promote equitable deployment of AI tools. At present, AI‐ECG primarily functions as a triage adjunct that may expedite diagnosis; however, definitive reductions in HF admissions or mortality await confirmation in prospective outcome trials. 3. AI in Cardiac Imaging: Enhancing Diagnostic Precision and Efficiency Cardiac imaging plays a pivotal role in the diagnosis and management of heart failure (HF), providing essential information about cardiac structure, function, and pathology [ 16 ]. Among imaging modalities, echocardiography is invaluable due to its non‐invasive nature, availability, and cost‐effectiveness [ 16 ]. However, its accuracy and efficiency can be compromised by operator dependency and the time required for image acquisition and interpretation [ 17 ]. As a result, there has been growing interest in the application of AI to streamline both the acquisition and interpretation of echocardiographic data, ultimately improving diagnostic precision and workflow efficiency. AI applications differ between HFrEF and HFpEF. HFrEF models focus on automated detection of reduced LVEF or overt systolic dysfunction, whereas HFpEF algorithms integrate diastolic indices, atrial strain, and Doppler‐based features to capture subtler functional impairment. A notable example of AI in echocardiography is EchoGo Heart Failure (Ultromics), a deep learning model that analyzes a single apical four‐chamber view to detect heart failure with HFpEF [ 18 ]. In initial testing, it achieved 87.8% sensitivity and 81.9% specificity for HFpEF diagnosis [ 18 ]. External validation on over 1000 patients at Beth Israel Deaconess confirmed comparable accuracy, while the related PanEcho model‐maintained performance across independent cohorts [ 19 ]. Patients identified as positive by the AI system were about twice as likely to experience adverse outcomes, underscoring its prognostic value [ 19 ]. AI in echocardiography is also revolutionizing the assessment of systolic function and chamber sizes. AI‐driven systems such as EchoNet‐Peds from Stanford have demonstrated the ability to automatically segment the LV and calculate EF in a matter of seconds with mean absolute error of 3.66%, matching the performance of expert human readers [ 20 ]. This capability is not limited to research settings; real‐world AI tools are now being integrated into commercial echocardiography platforms. For example, systems from GE Healthcare, Siemens Medical Solutions Inc., TOMTEC Imaging Systems GmbH, Ultromics Ltd., and Philips are equipped with machine learning algorithms that can automatically report EF, auto LV and LA volumes, auto Strain for manually selected views, further reducing operator burden and enhancing diagnostic consistency [ 21 , 22 ]. A major recent advancement is PanEcho , an AI system developed to automate echocardiographic interpretation. Trained on 1.2 million videos from 32,265 transthoracic echocardiograms (TTEs) across 24,405 patients, it demonstrated strong internal performance with a median AUC of 0.91 across 18 diagnostic tasks and a median mean absolute error (MAE) of 0.13 for 21 quantitative parameters. The model estimated left ventricular ejection fraction with an MAE of 4.2% (internal) and 4.5% (external), detected moderate or greater LV systolic dysfunction with AUCs of 0.98 and 0.99, and identified severe aortic stenosis with AUCs of 0.98 and 1.00, respectively. PanEcho sustained high accuracy when applied to abbreviated TTEs and point‐of‐care ultrasound, achieving median AUCs of 0.91 and 0.85 for 15 and 14 diagnostic tasks, respectively. This technology enables near‐instant, comprehensive echocardiographic reports, representing a substantial step forward in diagnostic efficiency [ 23 ]. Cardiac magnetic resonance imaging (CMR) remains the gold standard for assessing cardiac volumes, scar burden, and tissue characteristics. AI now addresses long‐standing challenges in post‐processing, particularly in ventricular and scar segmentation [ 24 ]. Deep learning methods, especially CNNs have markedly accelerated CMR analysis [ 25 ]. A CNN developed in England for landmark detection across cine, late gadolinium enhancement (LGE), and T1‐mapping sequences was trained on 2,329 patients and tested on 531. Detection accuracy on long‐axis images ranged from 99.7% to 100% for cine and 99.2%–99.5% for LGE; on short‐axis images, 96.6% for cine, 97.6% for LGE, and 98.7% for T1 mapping. Model outputs closely matched manual labels, with Euclidean deviations of 2–3.5 mm comparable to inter‐reader variation, supporting reliable automation in clinical CMR [ 25 ]. Commercial platforms such as Circle CVI and Arterys have integrated AI to enhance workflow speed and precision [ 26 , 27 ]. A UK study evaluating AI‐based biventricular volumetric analysis found agreement with manual contours, with greater consistency for left‐ventricular than right‐ventricular measures. Manual refinement further improved accuracy and reduced variability in ejection‐fraction estimates. Analysis time fell dramatically from 250 min to 6 min though clinicians noted residual need for oversight [ 26 ]. While full clinical adoption awaits further validation, AI‐enabled CMR clearly mitigates analytic bottlenecks, particularly in high‐volume heart‐failure assessment. Beyond HFpEF and systolic function assessment, AI is advancing the detection of valvular heart disease. Algorithms applied to echocardiography and even ECGs have successfully identified severe valve lesions [ 27 ]. Kwon et al. developed a deep‐learning model using 39,371 ECGs and patient demographics to detect moderate‐to‐severe aortic stenosis (AS), achieving an AUC of 0.86 with 80% sensitivity and 78.3% specificity on external validation. Although its positive‐predictive value (PPV) was low (10%), the negative‐predictive value (NPV) reached 99%, indicating it rarely missed significant AS. The same algorithm was adapted for mitral regurgitation (MR), yielding 90% sensitivity, 67% specificity, and a 99.4% NPV, again with a 10% PPV [ 28 ]. Additionally, AI‐enhanced chest X‐rays and point‐of‐care ultrasound are emerging as useful screening tools for heart failure, capable of detecting pulmonary edema and related pathologies [ 29 ]. Over 40 AI‐based devices hold regulatory clearance, each for defined, narrow tasks such as EF estimation or image segmentation rather than broad HF management. These authorizations apply to narrow diagnostic tasks rather than comprehensive HF management, and human over‐read remains required [ 30 ]. Additionally, these tools have been validated across multi‐center datasets, further supporting their widespread applicability and effectiveness [ 30 ]. As AI continues to evolve, it is directed to transform cardiac imaging, making it faster, more accurate, and more accessible, while also enabling early detection and personalized management of heart failure. These validation studies were largely single‐center with limited follow‐up, which constrains reproducibility despite promising internal metrics. Beyond generic HF phenotyping, AI‐enhanced imaging has shown promise in identifying specific etiologies such as hypertrophic cardiomyopathy, amyloidosis, and sarcoidosis through texture‐ and feature‐based analysis, though this remains outside the scope of the present review. 4. Remote Monitoring and Wearables: Harnessing AI for Early Detection of Heart Failure Continuous monitoring of physiological signals offers a powerful tool for detecting HF decompensation earlier than traditional sporadic clinic visits. With the advent of wearable devices, home monitoring technologies, and implantable sensors, healthcare providers are now able to track patients' physiological status in real‐time, providing critical insights into their condition. These devices, when combined with AI, these tools can analyze large amounts of data, identify early warning signs, and predict the risk of worsening heart failure, enabling timely interventions and potentially avoiding or reducing hospitalizations. The landmark LINK‐HF trial ( Multisensor Non‐invasive Remote Monitoring for Prediction of Heart Failure Exacerbation ) demonstrated the potential of AI‐enabled remote monitoring [ 31 ]. Using a wearable chest patch that continuously captured ECG, respiratory, and motion data, a personalized machine‐learning model established each patient's physiological baseline and detected deviations indicative of decompensation. Over 3 months, the system predicted impending heart‐failure hospitalizations with 76%–88% sensitivity and 85% specificity, providing a mean early warning of 6.5 days; performance comparable to implantable hemodynamic monitors used in clinical practice [ 31 ]. Personalization of models to individual baselines represents a key advance in predictive care, enabling timely, patient‐specific interventions. Effective integration of AI‐derived alerts into structured clinical workflows with defined escalation and accountability remains essential to mitigate alert fatigue and ensure clinical utility. Beyond wearables, implantable devices increasingly integrate AI to enhance heart‐failure monitoring. Boston Scientific's HeartLogic index derives a composite score from sensors embedded in ICD/CRT leads that capture heart sounds, thoracic impedance, respiration, and activity [ 32 ]. The MultiSENSE study validated this algorithm in 900 patients followed for 1 year, demonstrating 70% sensitivity for detecting HF events and providing a median 34‐day lead time before decompensation, offering clinicians an effective early‐warning tool [ 33 ]. Similarly, Abbott's CardioMEMS pulmonary‐artery pressure sensor applies AI analytics to guide diuretic titration by interpreting pressure trends [ 34 ]. In LVAD patients, hemodynamic‐guided management using CardioMEMS significantly lowered pulmonary‐artery diastolic pressure (PAD) in responders, correlating with improved 6‐min‐walk distance and reduced HF hospitalizations (12% vs 38.9%, p = 0.005) among those maintaining PAD < 20 mmHg [ 35 ]. As wearable technology continues to evolve, devices including smartwatches, rings, or even clothing equipped with sensors capable of tracking important physiological metrics [ 36 ]. Consumer devices such as Fitbit, Apple Watch, and Oura already monitor parameters like heart rate, heart rate variability, and physical activity [ 37 ]. Researchers are currently studying how AI algorithms applied to the data collected by these devices could predict heart failure events. For instance, Pilot systems, such as MIT's CHAIS (Cardiac Health AI System), have demonstrated the ability to estimate patient hemodynamics using just a single‐lead ECG patch in combination with deep learning algorithms [ 38 ]. Early data suggests that such AI‐based systems could noninvasively estimate left atrial pressure, which is crucial for assessing heart failure, and help identify when invasive procedures like catheterization may be needed [ 38 ]. An important aspect of integrating AI‐based monitoring systems into clinical practice is the incorporation of these AI alerts into existing care pathways. In the LINK‐HF2 trial, 95% of AI‐generated alerts were reviewed by clinicians within 24 h, and 26.7% of these alerts led to clinical action, such as adjusting therapy [ 39 ]. Qualitative feedback from healthcare providers in these pilot studies emphasized that trust in AI outputs and integration with electronic health EHRs are critical factors for widespread adoption [ 39 ]. While the field is still in its early stages, the initial evidence suggests that the combination of multivariate telemetry and machine learning has the potential to provide actionable insights that can improve patient care [ 40 ]. Table 1 compares conventional and AI‐enabled methods for HF diagnosis and prognosis, summarizing their respective advantages, limitations, and levels of clinical readiness. Table 1. Comparison of conventional and AI‐enabled approaches in heart‐failure diagnosis and management. Domain Conventional approach AI‐enabled approach Clinical readiness ECG screening Visual/manual interpretation Deep‐learning ECG detection of LV dysfunction FDA‐cleared, early clinical use Echocardiography/CMR Manual measurement Automated EF and diastolic‐function quantification Integrated into commercial software Remote monitoring Periodic clinic visits Multisensor + ML early decompensation alerts Pilot/registry validation Therapy titration Rule‐based, clinician‐only AI‐guided GDMT titration platforms Experimental, pilot trials Open in a new tab Table 2 provide a comparative overview of commonly used AI algorithms in heart‐failure diagnostics, imaging interpretation, and remote monitoring. This summary highlights key strengths and limitations of major architectures discussed in the manuscript and illustrates how different model classes contribute to the expanding AI toolkit for HF management. Table 2. Comparison of common artificial‐intelligence algorithms in heart‐failure diagnosis and management. Algorithm type Primary applications in HF Strengths Limitations Convolutional neural networks (CNNs) ECG interpretation, EF estimation, image segmentation Excellent spatial feature extraction; high performance on ECG and echocardiography; fast inference Limited capture of long‐range temporal dependencies; requires large labeled datasets Recurrent neural networks (RNNs)/LSTMs Time‐series signals from remote monitoring and telemetry Effective for sequential physiologic data; useful for early decompensation prediction Training instability; largely replaced by transformer models Transformers ECG interpretation, multimodal fusion, prognostic modeling Capture long‐range temporal patterns; scalable and high performing with large datasets Computationally intensive; require substantial training data Multimodal DL models Integration of ECG, imaging, labs, and wearable data Better performance in multimorbidity; robust across data types; reduces single‐modality bias Architecturally complex; limited external validation; computationally demanding Gradient boosting/XGBoost Risk stratification, readmission prediction, mortality modeling Strong performance on tabular data; interpretable; low computational cost Cannot process raw ECG/imaging data; relies on engineered features Logistic regression/Classical ML Baseline clinical prediction models Transparent, simple, widely understood Lower accuracy than deep learning on physiologic signals; limited capacity for complex patterns Open in a new tab 5. Smart Devices and Point‐of‐Care AI in Heart Failure Management AI is increasingly embedded in smart medical devices for rapid, point‐of‐care heart failure (HF) diagnostics [ 41 ]. A leading example is the convolutional neural network (CNN)–based digital stethoscope developed by Eko Health , which integrates ECG recording and AI analytics [ 42 ]. The device captures a 15‐second heart‐sound and ECG trace, which the algorithm instantly analyzes to flag potential left‐ventricular dysfunction. In a recent multicentre prospective study, the AI‐enabled stethoscope accurately detected reduced LVEF (≤ 40%). When placed at the pulmonary valve position, it achieved an AUROC of 0.85, with 84.8% sensitivity and 69.5% specificity. Combining pulmonary and handheld positions yielded similar AUROC (0.85) with 82.7% sensitivity and 79.9% specificity. A weighted logistic regression model further improved performance (AUROC 0.91; sensitivity 91.9%; specificity 80.2%) [ 42 ]. Such real‐time analysis enables frontline clinicians to rapidly identify patients at risk of HFrEF and expedite confirmatory echocardiography. Consumer‐grade wearables now play an expanding role in HF management. The FDA‐cleared Apple Watch (Series 4 and newer) and AliveCor Kardia devices exemplify how AI‐enhanced wearables enable proactive rhythm surveillance [ 43 ]. In a comparative study of 200 participants (162 sinus rhythm, 38 atrial fibrillation [AF]), the Apple Watch Series 4 achieved 100% accuracy for sinus rhythm and 90.5% for AF, while KardiaMobile reached 99.0% and 100%, respectively. For heart‐rate measurement during sinus rhythm, KardiaMobile achieved 94.4% accuracy, versus 90.7% with the Apple Watch's photoplethysmography (PPG) and 96.3% using its ECG mode; in AF, accuracies were 91.3%, 82.6%, and 87.0%, respectively [ 43 ]. Because AF worsens HF outcomes, early rhythm detection through such devices facilitates timely initiation of anticoagulation or rate‐control therapy [ 44 ]. The cardiology IoT ecosystem continues to broaden with innovations such as “smart socks,” which detect peripheral edema via embedded sensors [ 45 ]. These devices generate continuous data streams amenable to AI analysis, expanding opportunities for real‐time HF monitoring and intervention. 6. Clinical Relevance and Implementation For practicing cardiologists, the important question is how AI‐driven innovations translate into tangible improvement in patient care. Early evidence is promising. AI‐assisted ECG analysis can function as a “digital screening” tool by flagging patients who may benefit from timely echocardiography or therapy adjustment [ 46 ]. Automated echocardiographic analysis reduces reporting delays and inter‐observer variability, allowing more rapid patient triage [ 21 ]. Remote AI monitoring platforms offer the potential to decrease hospital readmissions by detecting early signs of congestion and enabling proactive intervention before symptoms worsen [ 47 ]. Devices such as Eko's AI‐enabled stethoscope extend heart failure screening capabilities beyond specialized cardiology centers to primary care settings, increasing accessibility [ 42 ]. Interoperability can be achieved through adoption of HL7‐FHIR, DICOM‐SR, and SMART‐on‐FHIR interfaces linking AI outputs to existing EHRs. Preliminary evidence from CardioMEMS and LINK‐HF indicates potential reductions in readmissions, though comprehensive cost‐effectiveness data remain lacking. Beyond imaging and signal analysis, large language models (LLMs) are being explored for HF patient education, automated triage messaging, and rapid literature summarization. Such tools remain experimental but highlight the growing role of generative AI in patient‐clinician communication. AI‐enabled decision‐support platforms are increasingly used for guideline‐directed medical‐therapy (GDMT) titration and remote management. By integrating vital‐sign trends, device data, and prior medication responses, these systems assist clinicians in adjusting doses and diuretics in real time. Early studies suggest improved timeliness of therapy changes, though validation in large pragmatic trials is pending. Emerging AI frameworks also model drug‐response curves and side‐effect probabilities to support individualized GDMT optimization. Such systems remain investigational but illustrate the potential of precision‐pharmacotherapy guided by real‐world data. Despite rapid progress, cautious implementation and structured clinician training remain critical. Clinicians must recognize AI's limitations, including false positives and negatives, and interpret outputs as decision‐support tools rather than substitutes for clinical judgment [ 47 ]. Regulatory and reimbursement frameworks are evolving in parallel; the FDA has already authorized multiple AI‐ and machine‐learning–based medical devices [ 48 ]. Prospective clinical trials and health‐economic evaluations are still required to confirm that AI adoption improves outcomes and cost‐effectiveness, evidence essential for broader acceptance in routine practice. Comprehensive AI‐literacy and workflow‐training programs for cardiology teams are indispensable to ensure safe and effective clinical integration. It is be worth discussing that symptoms and physiologic abnormalities in COPD, renal impairment, and liver disease can mimic HF, multimorbidity remains a key source of diagnostic error. Emerging multimodal AI frameworks that integrate ECG, imaging, laboratory markers, and longitudinal trajectories may help differentiate HF from non‐cardiac conditions. However, such systems still require clinician oversight, as comorbidity‐related changes can influence model outputs and must be interpreted within the broader differential diagnosis. 7. Challenges, Limitations, and Ethical Considerations in AI‐Driven HF Management Despite its transformative potential, the integration of AI into HF management introduces technical, ethical, and regulatory challenges all of which warrant careful scrutiny. Many high‐performing AI models are trained on single‐center or demographically narrow datasets, often overrepresenting white or male patients. Consequently, model performance and clinical applicability decline in more diverse populations. A study in cardiovascular imaging demonstrated substantial variation in algorithm accuracy by race and sex, potentially leading to unequal care outcomes [ 49 ]. Data heterogeneity further compounds this issue: real‐world clinical data are fragmented, inconsistently labeled, and collected using variable equipment and protocols. Even minor ECG variations, such as noise or lead placement, can alter AI predictions [ 50 ]. Such variability increases false‐positive alerts, emphasizing the need for threshold optimization and local recalibration. Algorithmic bias can arise from unbalanced training data or outcome labeling, reinforcing existing disparities even without explicit demographic inputs [ 51 ]. Overfitting remains a persistent risk, models that perform well on training data may fail in external cohorts, particularly given the phenotypic diversity of HF (e.g., HFpEF vs. HFrEF) [ 52 ]. Robust external validation across geographically and demographically distinct populations remains uncommon, undermining reliability [ 53 ]. Mandatory subgroup‐specific reporting and periodic recalibration are essential to mitigate bias and maintain fairness. Deep‐learning architectures such as CNNs and transformers often function as “black boxes,” limiting interpretability and clinician trust [ 54 ]. Methods like SHAP (SHapley Additive exPlanations) and Grad‐CAM (Gradient‐weighted Class Activation Mapping) provide partial insights into decision pathways but are not yet routinely integrated into practice [ 55 ]. Lack of transparency complicates clinical adoption and regulatory approval. AI integration into wearables and remote monitoring introduces heightened concerns over data privacy and patient autonomy. These systems generate large volumes of sensitive data that require strong governance, anonymization, and transparent consent frameworks [ 56 ]. Privacy can be protected through encrypted transmission, local edge processing, and federated‐learning architectures that retain raw data on‐device. Determining responsibility for AI‐related diagnostic or therapeutic errors remains legally ambiguous [ 57 ]. When a system misclassifies a patient—for example, labeling high‐risk HF as low‐risk, it is unclear whether liability rests with the clinician, software vendor, or institution. Current laws generally hold physicians accountable for exercising clinical judgment [ 58 ]. However, shared‐liability frameworks are emerging, advocating for collective accountability proportional to the degree of human oversight [ 59 ]. Clear documentation of AI involvement, validation transparency, and explicit inclusion of algorithmic contributions in informed‐consent processes will be crucial as AI becomes more embedded in clinical workflows. Besides that, this review does not provide detailed evaluation of AI‐based differentiation of HFrEF, HFmrEF, and HFpEF or prediction of transitions between these phenotypes, which represents an important but separate area requiring dedicated investigation. 8. Future Directions As artificial intelligence continues to evolve, its integration into heart‐failure care will increasingly emphasize explainability, interoperability, and equity. Future research should prioritize multi‐center, multi‐ethnic validation cohorts to ensure generalizability and mitigate algorithmic bias. Federated‐learning frameworks may enable collaborative model development without compromising patient privacy. Emerging work is also exploring AI‐enabled physiologic sensing in mechanical circulatory support systems and personalized artificial‐heart platforms; however, these device‐level surgical technologies remain outside the scope of this review and require dedicated evaluation as evidence matures. Future work may also explore AI‐driven classification of early acute, early post‐implant, and late post‐LVAD right‐heart failure, although these device‐specific surgical phenotypes fall outside the scope of validated AI applications reviewed here. 9. Conclusion This review highlights how artificial intelligence is fundamentally transforming HF management by addressing persistent challenges in diagnosis, risk stratification, and therapy personalization. AI‐powered tools have demonstrated the ability to detect both systolic and diastolic dysfunction from routine ECGs, often identifying disease earlier than conventional methods. In cardiac imaging, deep learning models are streamlining echocardiographic and CMR interpretation with accuracy approaching that of expert clinicians, while also reducing workflow burden. Remote monitoring systems and wearable devices, integrated with machine learning algorithms, offer continuous, real‐time insights that support proactive management and reduce avoidable hospitalizations. Importantly, these technologies are broadening access to advanced HF care by bringing sophisticated diagnostics into primary care and community settings. However, challenges remain. Issues such as generalizability across populations, data heterogeneity, algorithmic bias, interpretability, liability, and integration with electronic health records must be carefully addressed to ensure equitable and safe implementation. Despite these hurdles, the growing body of evidence supports AI not merely as a supportive tool but as a foundational component of next‐generation HF care. With ongoing validation, ethical oversight, and thoughtful integration, AI holds the potential to significantly enhance outcomes, personalize therapy, and improve quality of life for patients worldwide. All systems reviewed are clinician‐supervised decision‐support tools rather than autonomous diagnostic agents. Author Contributions Farrukh Ansar: conceptualization, methodology, data curation, investigation, formal analysis, visualization, project administration, writing – original draft preparation, and writing – review and editing. Muhammad Aamir Waheed: data curation, investigation, validation, funding acquisition, project administration, resources, and writing – review and editing. Usman Zafar: conceptualization, investigation, formal analysis, project administration, writing – original draft preparation, and writing – review and editing. Abdulrahman Kolapo: conceptualization, methodology, validation, project administration, resources, and writing – review and editing. Walid Sarfaraz: methodology, data curation, investigation, validation, resources, and writing – review and editing. Khalid Rashid: software, data curation, investigation, validation, supervision, project administration, writing – original draft preparation, and writing – review and editing. Conflicts of Interest The authors declare no conflict of Interest. Transparency Statement The lead author Muhammad Aamir Waheed affirms that this manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained. Acknowledgments Hamad Medical Corporation funds the study. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request. (This article is a narrative review based entirely on previously published studies. All data supporting the findings of this study are available within the cited literature and its supporting materials). References 1. Savarese G., Becher P. M., Lund L. H., Seferovic P., Rosano G. M. C., and Coats A. J. S., “Global Burden of Heart Failure: A Comprehensive and Updated Review of Epidemiology,” Cardiovascular Research 118, no. 17 (2023): 3272–3287, 10.1093/cvr/cvac013. [ DOI ] [ PubMed ] [ Google Scholar ] 2. Fragasso G., “The Concept of “Heart Failure With Preserved Ejection Fraction”: Time for a Critical Reappraisal,” Reviews in Cardiovascular Medicine 24, no. 7 (2023): 202, 10.31083/j.rcm2407202. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 3. Formiga F., Nuñez J., Castillo Moraga M. J., et al., “Diagnosis of Heart Failure With Preserved Ejection Fraction: A Systematic Narrative Review of the Evidence,” Heart Failure Reviews 29, no. 1 (2024): 179–189, 10.1007/s10741-023-10360-z. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 4. Bastos J. M., Colaço B., Baptista R., Gavina C., and Vitorino R., “Innovations in Heart Failure Management: The Role of Cutting‐Edge Biomarkers and Multi‐Omics Integration,” Journal of Molecular and Cellular Cardiology Plus 11 (2025): 100290, 10.1016/j.jmccpl.2025.10029. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Hughes J. W., Yuan N., He B., et al., “Deep Learning Evaluation of Biomarkers From Echocardiogram Videos,” EBioMedicine 73 (2021): 103613, 10.1016/j.ebiom.2021.103613. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 6. Sutton R. T., Pincock D., Baumgart D. C., Sadowski D. C., Fedorak R. N., and Kroeker K. I., “An Overview of Clinical Decision Support Systems: Benefits, Risks, and Strategies for Success,” npj Digital Medicine 3 (2020): 17, 10.1038/s41746-020-0221-y. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 7. Maleki Varnosfaderani S. and Forouzanfar M., “The Role of AI in Hospitals and Clinics: Transforming Healthcare in the 21st Century,” Bioengineering 11, no. 4 (2024): 337, 10.3390/bioengineering11040337. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 8. Medhi D., Kamidi S. R., Mamatha Sree K. P., et al., “Artificial Intelligence and its Role in Diagnosing Heart Failure: A Narrative Review,” Cureus 16, no. 5 (2024): e59661, 10.7759/cureus.59661. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 9. Cersosimo A., Zito E., Pierucci N., Matteucci A., and La Fazia V. M., “A Talk With ChatGPT: The Role of Artificial Intelligence in Shaping the Future of Cardiology and Electrophysiology,” Journal of Personalized Medicine 15, no. 5 (2025): 205, 10.3390/jpm15050205. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 10. Sattar Y. and Chhabra L. Electrocardiogram. [Updated 2023 June 5]. In: StatPearls. Treasure Island (FL): StatPearls Publishing. https://www.ncbi.nlm.nih.gov/books/NBK549803/ . 11. Shemisa K., Bhatt A., Cheeran D., and Neeland I. J., “Novel Biomarkers of Subclinical Cardiac Dysfunction in the General Population,” Current Heart Failure Reports 14, no. 4 (2017): 301–310, 10.1007/s11897-017-0342-z. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 12. Martínez‐Sellés M. and Marina‐Breysse M., “Current and Future Use of Artificial Intelligence in Electrocardiography,” Journal of Cardiovascular Development and Disease 10, no. 4 (2023): 175, 10.3390/jcdd10040175. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 13. Attia Z. I., Kapa S., Yao X., et al., “Prospective Validation of a Deep Learning Electrocardiogram Algorithm for the Detection of Left Ventricular Systolic Dysfunction,” Journal of Cardiovascular Electrophysiology 30, no. 5 (2019): 668–674, 10.1111/jce.13889. [ DOI ] [ PubMed ] [ Google Scholar ] 14. Lee E., Ito S., Miranda W. R., et al., “Artificial Intelligence‐Enabled ECG for Left Ventricular Diastolic Function and Filling Pressure,” npj Digital Medicine 7, no. 1 (2024): 4, 10.1038/s41746-023-00993-7. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 15. Dhingra L. S., Aminorroaya A., Sangha V., et al., “Heart Failure Risk Stratification Using Artificial Intelligence Applied to Electrocardiogram Images: A Multinational Study,” European Heart Journal 46, no. 11 (2025): 1044–1053, 10.1093/eurheartj/ehae914. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Kim S. R. and Park S. M., “Role of Cardiac Imaging in Management of Heart Failure,” Korean Journal of Internal Medicine 38, no. 5 (2023): 607–619, 10.3904/kjim.2023.262. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 17. Yu C. M., “Challenges and Opportunity in the Era of Quantitative Echocardiography,” Echo Research and Practice 4, no. 4 (2017): E3–E6, 10.1530/ERP-17-0049. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 18. Akerman A. P., Porumb M., Scott C. G., et al., “Automated Echocardiographic Detection of Heart Failure With Preserved Ejection Fraction Using Artificial Intelligence,” JACC: Advances 2, no. 6 (2023): 100452, 10.1016/j.jacadv.2023.100452. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 19. Akerman A. P., Al‐Roub N., Angell‐James C., et al., “External Validation of Artificial Intelligence for Detection of Heart Failure With Preserved Ejection Fraction,” Nature Communications 16, no. 1 (2025): 2915, 10.1038/s41467-025-58283-7. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 20. Reddy C. D., Lopez L., Ouyang D., Zou J. Y., and He B., “Video‐Based Deep Learning for Automated Assessment of Left Ventricular Ejection Fraction in Pediatric Patients,” Journal of the American Society of Echocardiography 36, no. 5 (2023): 482–489, 10.1016/j.echo.2023.01.015. [ DOI ] [ PubMed ] [ Google Scholar ] 21. Hirata Y. and Kusunose K., “AI in Echocardiography: State‐of‐the‐Art Automated Measurement Techniques and Clinical Applications,” JMA Journal 8, no. 1 (2025): 141–150, 10.31662/jmaj.2024-0180. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Shiokawa N., Izumo M., Shimamura T., et al., “Accuracy and Efficacy of Artificial Intelligence‐Derived Automatic Measurements of Transthoracic Echocardiography in Routine Clinical Practice,” Journal of Clinical Medicine 13, no. 7 (2024): 1861, 10.3390/jcm13071861. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 23. Holste G., Oikonomou E. K., Tokodi M., Kovács A., Wang Z., and Khera R., “PanEcho: Complete AI‐Enabled Echocardiography Interpretation With Multi‐Task Deep Learning,” medRxiv [Preprint] (2025). 10.1101/2024.11.16.24317431. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 24. Aromiwura A. A., Cavalcante J. L., Kwong R. Y., et al., “The Role of Artificial Intelligence in Cardiovascular Magnetic Resonance Imaging,” Progress in Cardiovascular Diseases 86 (2024): 13–25, 10.1016/j.pcad.2024.06.004. [ DOI ] [ PubMed ] [ Google Scholar ] 25. Xue H., Artico J., Fontana M., Moon J. C., Davies R. H., and Kellman P., “Landmark Detection in Cardiac MRI by Using a Convolutional Neural Network,” Radiology: Artificial Intelligence 3, no. 5 (2021): e200197, 10.1148/ryai.2021200197. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 26. Hatipoglu S., Mohiaddin R. H., Gatehouse P., et al., “Performance of Artificial Intelligence for Biventricular Cardiovascular Magnetic Resonance Volumetric Analysis in the Clinical Setting,” International Journal of Cardiovascular Imaging 38, no. 11 (2022): 2413–2424, 10.1007/s10554-022-02649-1. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 27. Nedadur R., Wang B., and Tsang W., “Artificial Intelligence for the Echocardiographic Assessment of Valvular Heart Disease,” Heart 108, no. 20 (2022): 1592–1599, 10.1136/heartjnl-2021-319725. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 28. Kwon J., Kim K. H., Akkus Z., Jeon K. H., Park J., and Oh B. H., “Artificial Intelligence for Detecting Mitral Regurgitation Using Electrocardiography,” Journal of Electrocardiology 59 (2020): 151–157, 10.1016/j.jelectrocard.2020.02.008. [ DOI ] [ PubMed ] [ Google Scholar ] 29. Garza‐Frias E., Kaviani P., Karout L., et al., “Early Detection of Heart Failure With Autonomous AI‐Based Model Using Chest Radiographs: A Multicenter Study,” Diagnostics 14, no. 15 (2024): 1635, 10.3390/diagnostics14151635. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 30. Gerke S., Minssen T., and Cohen G., “Ethical and Legal Challenges of Artificial Intelligence‐Driven Healthcare,” Artificial Intelligence in Healthcare 1, (2020): 295–336, 10.1016/B978-0-12-818438-7.00012-5. [ DOI ] [ Google Scholar ] 31. Stehlik J., Schmalfuss C., Bozkurt B., et al., “Continuous Wearable Monitoring Analytics Predict Heart Failure Hospitalization: The LINK‐HF Multicenter Study,” Circulation: Heart Failure 13, no. 3 (2020): e006513, 10.1161/CIRCHEARTFAILURE.119.006513. [ DOI ] [ PubMed ] [ Google Scholar ] 32. Capucci A., Santini L., Favale S., et al., “Preliminary Experience With the Multisensor Heartlogic Algorithm for Heart Failure Monitoring: A Retrospective Case Series Report,” ESC Heart Failure 6, no. 2 (2019): 308–318, 10.1002/ehf2.12394. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 33. Boehmer J. P., Hariharan R., Devecchi F. G., et al., “A Multisensor Algorithm Predicts Heart Failure Events in Patients With Implanted Devices,” JACC: Heart Failure 5, no. 3 (2017): 216–225, 10.1016/j.jchf.2016.12.011. [ DOI ] [ PubMed ] [ Google Scholar ] 34. Thohan V., Abraham J., Burdorf A., et al., “INTELLECT 2‐HF Investigators. Use of a Pulmonary Artery Pressure Sensor to Manage Patients With Left Ventricular Assist Devices,” Circulation‐Heart Failure 16, no. 6 (2023): e009960, 10.1161/CIRCHEARTFAILURE.122.009960. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 35. Tolu‐Akinnawo O., Akhtar N., Zalavadia N., and Guglin M., “CardioMEMS Heart Failure System: An Up‐to‐Date Review,” Cureus 17, no. 1 (2025): e77816, 10.7759/cureus.77816. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 36. Guk K., Han G., Lim J., et al., “Evolution of Wearable Devices With Real‐Time Disease Monitoring for Personalized Healthcare,” Nanomaterials 9,no. 6 (2019): 813, 10.3390/nano9060813. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 37. Li K., Cardoso C., Moctezuma‐Ramirez A., Elgalad A., and Perin E., “Heart Rate Variability Measurement Through a Smart Wearable Device: Another Breakthrough for Personal Health Monitoring?,” International Journal of Environmental Research and Public Health 20, no. 24 (2023): 7146, 10.3390/ijerph20247146. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 38. Schlesinger D. E., Alam R., Ringel R., et al., “Artificial Intelligence for Hemodynamic Monitoring With a Wearable Electrocardiogram Monitor,” Communications Medicine 5, no. 1 (2025): 4, 10.1038/s43856-024-00730-5. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 39. Sideris K., Weir C. R., Schmalfuss C., et al., “Artificial Intelligence Predictive Analytics in Heart Failure: Results of the Pilot Phase of a Pragmatic Randomized Clinical Trial,” Journal of the American Medical Informatics Association 31, no. 4 (2024): 919–928, 10.1093/jamia/ocae017. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 40. Johnson A. E. W., Ghassemi M. M., Nemati S., Niehaus K. E., Clifton D., and Clifford G. D., “Machine Learning and Decision Support in Critical Care,” Proceedings of the IEEE 104, no. 2 (2016): 444–466. 10.1109/JPROC.2015.2501978. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 41. Yammouri G. and Ait Lahcen A., “AI‐Reinforced Wearable Sensors and Intelligent Point‐of‐Care Tests,” Journal of Personalized Medicine 14, no. 11 (2024): 1088, 10.3390/jpm14111088. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 42. Bachtiger P., Petri C. F., Scott F. E., et al., “Point‐of‐Care Screening for Heart Failure With Reduced Ejection Fraction Using Artificial Intelligence During Ecg‐Enabled Stethoscope Examination in London, UK: A Prospective, Observational, Multicentre Study,” Lancet Digital Health 4, no. 2 (2022): e117–e125, 10.1016/S2589-7500(21)00256-9. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 43. Sarkar M., Lee T. H., and Sahoo P. K., “Smart Healthcare: Exploring the Internet of Medical Things With Ambient Intelligence,” Electronics 13, no. 12 (2024. Jun): 2309, 10.3390/electronics13122309. [ DOI ] [ Google Scholar ] 44. Mogensen U. M., Jhund P. S., Abraham W. T., et al., “Type of Atrial Fibrillation and Outcomes in Patients With Heart Failure and Reduced Ejection Fraction,” Journal of the American College of Cardiology 70, no. 20 (2017): 2490–2500, 10.1016/j.jacc.2017.09.027. [ DOI ] [ PubMed ] [ Google Scholar ] 45. Fallahzadeh R., Pedram M., and Ghasemzadeh H., “SmartSock: A Wearable Platform for Context‐Aware Assessment of Ankle Edema.” Annu Int Conf (IEEE Eng Med Biol Soc, 2016. August). 2016, 6302–6306. 10.1109/EMBC.2016.7592169. [ DOI ] [ PubMed ] [ Google Scholar ] 46. Karatzia L., Aung N., and Aksentijevic D., “Artificial Intelligence in Cardiology: Hope for the Future and Power for the Present,” Frontiers in Cardiovascular Medicine 9 (2022): 945726, 10.3389/fcvm.2022.945726. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 47. Dubey A. and Tiwari A., “Artificial Intelligence and Remote Patient Monitoring in US Healthcare Market: A Literature Review,” Journal of Market Access and Health Policy 11, no. 1 (2023): 2205618, 10.1080/20016689.2023.2205618. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 48. Zhou K. and Gattinger G., “The Evolving Regulatory Paradigm of AI in MedtTech: A Review of Perspectives and Where We Are Today,” Therapeutic Innovation and Regulatory Science 58, no. 3 (2024): 456–464, 10.1007/s43441-024-00628-3. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 49. Seyyed‐Kalantari L., Zhang H., McDermott M. B. A., Chen I. Y., and Ghassemi M., “Underdiagnosis Bias of Artificial Intelligence Algorithms Applied to Chest Radiographs in Under‐Served Patient Populations,” Nature Medicine 27 (2021): 2176–2182, 10.1038/s41591-021-01595-0. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 50. Hannun A. Y., Rajpurkar P., Haghpanahi M., et al., “Cardiologist‐Level Arrhythmia Detection and Classification in Ambulatory Electrocardiograms Using a Deep Neural Network,” Nature Medicine 25 (2019): 65–69, 10.1038/s41591-018-0268-3. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 51. Obermeyer Z., Powers B., Vogeli C., and Mullainathan S., “Dissecting Racial Bias in an Algorithm Used to Manage the Health of Populations,” Science 366 (2019): 447–453, 10.1126/science.aax2342. [ DOI ] [ PubMed ] [ Google Scholar ] 52. Aliferis C. and Simon G., “Overfitting, Underfitting and General Model Overconfidence and Under‐Performance Pitfalls and Best Practices in Machine Learning and AI.,” in Artificial Intelligence and Machine Learning in Health Care and Medical Sciences: Best Practices and Pitfalls. Springer, 2024). https://www.ncbi.nlm.nih.gov/books/NBK610560/ . [ PubMed ] [ Google Scholar ] 53. Wu E., Wu K., Daneshjou R., Ouyang D., Ho D. E., and Zou J., “How Medical AI Devices Are Evaluated: Limitations and Recommendations From an Analysis of FDA Approvals,” Nature Medicine 27 (2021): 582–584, 10.1038/s41591-021-01312-x. [ DOI ] [ PubMed ] [ Google Scholar ] 54. Marey A., Arjmand P., Alerab A. D. S., et al., “Explainability, Transparency and Black Box Challenges of AI in Radiology: Impact on Patient Care in Cardiovascular Radiology,” Egyptian Journal of Radiology and Nuclear Medicine 55 (2024): 183, 10.1186/s43055-024-01356-2. [ DOI ] [ Google Scholar ] 55. Allgaier J., Mulansky L., Draelos R. L., and Pryss R., “How Does the Model Make Predictions? A Systematic Literature Review on the Explainability Power of Machine Learning in Healthcare,” Artificial Intelligence in Medicine 143 (2023): 102616, 10.1016/j.artmed.2023.102616. [ DOI ] [ PubMed ] [ Google Scholar ] 56. Jaime F. J., Muñoz A., Rodríguez‐Gómez F., and Jerez‐Calero A., “Strengthening Privacy and Data Security in Biomedical Microelectromechanical Systems by IoT Communication Security and Protection in Smart Healthcare,” Sensors 23, no. 21 (2023): 8944, 10.3390/s23218944. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 57. Cestonaro C., Delicati A., Marcante B., Caenazzo L., and Tozzo P., “Defining Medical Liability When Artificial Intelligence is Applied on Diagnostic Algorithms: A Systematic Review,” Frontiers in Medicine 10 (2023): 1305756, 10.3389/fmed.2023.1305756. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 58. Cohen I. G. Slottje A. Artificial intelligence and the law of informed consent Research Handbook on Health, AI and the Law. CheltenhamSolaiman B. Cohen I. G. U. K. Edward Elgar Publishing Ltd; 2024. https://www.ncbi.nlm.nih.gov/books/NBK613199/ 10.4337/9781802205657.ch10. [ DOI ] [ PubMed ] [ Google Scholar ] 59. Smith H. and Fotheringham K., “Artificial Intelligence in Clinical Decision‐Making: Rethinking Liability,” Medical Law International 20, no. 2 (2020): 131–154, 10.1177/0968533220945766. [ DOI ] [ 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 that support the findings of this study are available from the corresponding author upon reasonable request. (This article is a narrative review based entirely on previously published studies. All data supporting the findings of this study are available within the cited literature and its supporting materials). Articles from Health Science Reports are provided here courtesy of Wiley ACTIONS View on publisher site PDF (736.4 KB) 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