多视角深度学习改善了超声心动图对主要心脏病的检测。
Barrios, J.P., Ansari, M.U., Olgin, J.E., Abreau, S., Delfrate, J., Langlais, E.L., Avram, R., & Tison, G.H. (2026). Multiview deep learning improves detection of major cardiac conditions from echocardiography.. Nat Cardiovasc Res. https://doi.org/10.1136/bmj.310.6973.170
Barrios JP, Ansari MU, Olgin JE, Abreau S, Delfrate J, Langlais EL, et al. Multiview deep learning improves detection of major cardiac conditions from echocardiography.. Nat Cardiovasc Res. 2026; doi: 10.1136/bmj.310.6973.170
Barrios JP, Ansari MU, Olgin JE, et al. Multiview deep learning improves detection of major cardiac conditions from echocardiography.[J]. Nat Cardiovasc Res. 2026. DOI: 10.1136/bmj.310.6973.170.
@article{barrios2026,
author = {Joshua P Barrios and Minhaj U Ansari and Jeffrey E Olgin and Sean Abreau and Jacques Delfrate and Elodie L Langlais and Robert Avram and Geoffrey H Tison},
title = {Multiview deep learning improves detection of major cardiac conditions from echocardiography.},
journal = {Nat Cardiovasc Res},
year = {2026},
doi = {10.1136/bmj.310.6973.170},
note = {PMID: 41844861},
}
TY - JOUR AU - Joshua P Barrios AU - Minhaj U Ansari AU - Jeffrey E Olgin AU - Sean Abreau AU - Jacques Delfrate AU - Elodie L Langlais AU - Robert Avram AU - Geoffrey H Tison TI - Multiview deep learning improves detection of major cardiac conditions from echocardiography. T2 - Nat Cardiovasc Res PY - 2026 DO - 10.1136/bmj.310.6973.170 AN - PMID:41844861 ER -
Medical imaging often captures multiple two-dimensional views of three-dimensional anatomic structures, but most artificial intelligence (AI) models analyze two-dimensional data. Here we show that integrating multiple imaging views using a single AI model can improve diagnostic performance. We developed a deep neural network (DNN) architecture that combines information from multiple video views simultaneously. Using echocardiogram data from the University of California, San Francisco, and the Montreal Heart Institute, we applied our multiview DNN approach for three primary demonstration tasks: detecting any left or right ventricular abnormality, diastolic dysfunction, and substantial valvular regurgitation. Across various tasks, our multiview DNNs improved discrimination as measured by the area under the receiver operating characteristic curve by 0.06-0.09 compared to DNNs trained on any single view. This demonstrates that AI models that can combine information from multiple imaging views simultaneously can better capture complex anatomy and physiology for certain tasks, underscoring the value of a multiview paradigm for AI in medical imaging.