A multimodal deep learning framework combining chest X-rays and electrocardiograms classified frailty in older adults more accurately than models relying on single imaging modalities or clinical baselines, researchers reported in PLOS Digital Health. The retrospective study evaluated whether combining routinely collected clinical imaging could offer an objective, rapid screening approach to identify patients who may require comprehensive geriatric assessment.
Frailty represents an age-related syndrome linked to elevated risks of disability, hospitalization, and mortality. While the standard Comprehensive Geriatric Assessment provides thorough clinical evaluation, it remains subjective, resource-intensive, and time-consuming. To address these limitations, investigators analyzed retrospective data from 2,830 older adults who underwent geriatric evaluation between 2007 and 2024, partitioning the cohort chronologically into training, validation, and test datasets.
The researchers defined frailty as a Multidimensional Frailty Score greater than 5 and compared three distinct neural network architectures: an EVA-X vision transformer trained solely on chest X-rays, a DenseNet-121 model trained exclusively on electrocardiograms, and a multimodal architecture integrating both imaging sources. The models were additionally tested against four non-imaging clinical baselines incorporating combinations of age, sex, comorbidity index scores, or routine laboratory measurements.
The combined multimodal model achieved the strongest classification performance, yielding an area under the receiver operating characteristic curve of 0.778 and an area under the precision-recall curve of 0.726. When estimating continuous Multidimensional Frailty Scores, the multimodal approach similarly outperformed alternative configurations, registering a Pearson correlation coefficient of r = 0.610 and a mean absolute error of 1.68.
Qualitative inspection with Gradient-weighted Class Activation Mapping highlighted specific image regions that influenced the system’s classifications. Although the authors noted that the architecture demonstrates modest classification performance and feasibility for rapid screening, they emphasized that further studies are necessary to evaluate its clinical utility for stratifying postoperative risk in preoperative settings.
Subject of Research: Deep learning frailty classification using electrocardiograms and chest radiography
Article Title: Multimodal deep learning approach for frailty classification using electrocardiogram and chest X-ray imaging
Article References: Park, J., Choi, J.-Y., & Kim, K.-I. (2026). Multimodal deep learning approach for frailty classification using electrocardiogram and chest X-ray imaging. PLOS Digital Health, 5(11), e0001838. https://doi.org/10.1371/journal.pdig.0001838
Image Credits: AI Generated
DOI: 10.1371/journal.pdig.0001838
Keywords: deep learning, multimodal AI, frailty screening, older adults, electrocardiogram, chest X-ray, comprehensive geriatric assessment, geriatric medicine, Multidimensional Frailty Score, EVA-X, DenseNet-121, AUROC, AUPRC, Grad-CAM, clinical imaging, preoperative risk stratification
News Source: Beatrice Stafford. (November 19, 2026). Multimodal AI Screens Older Adults for Frailty Using Routine Imaging. Scienmag.



