Mitral regurgitation detection and central/eccentric classification using transformer-based deep learning in multi-view echocardiography.
Fully automated diagnostic echocardiography allows for broader screening, earlier diagnosis for patients with mitral regurgitation (MR). This feasibility study developed a deep learning (DL) framework to automatically detect MR in echocardiography videos and classify regurgitation samples into central or eccentric regurgitation.
We designed a Transformer-based deep learning model for fully automatic Doppler video detection of MR. This framework can automatically detect MR and classify central or eccentric regurgitation. The algorithm was trained, validated, and tested using retrospectively selected studies. A prospective dataset of 217 patients was used as an independent test database.
(a) The model demonstrated exceptional diagnostic accuracy for MR, achieving accuracy rates of 0.94(95%CI: 0.90-0.97) and 0.92(95%CI: 0.89-0.96) in retrospective and prospective test sets, respectively, with area under the curve (AUC) values of 0.98 (95%CI: 0.96-1.0) and 0.97 (95%CI: 0.95-0.99). Its performance aligned with that of middle seniority physicians. (b) The model achieved an accuracy of 0.93 in identifying central and eccentric MR across both retrospective and prospective test sets, AUC values of 0.96 (95%CI: 0.89-0.99) and 0.95 (95%CI: 0.89-0.98), respectively. (c) On the tasks of MR, the diagnostic performance of the model in multiple views was better than that of the parasternal long-axis view (PLAX) alone or the apical four-chamber view (A4C) in both retrospective and prospective datasets.
The Transformer-based algorithm proposed in this study can automate and improve the efficiency of clinical workflows, screen multi-view for the presence of MR, and assist in classifying regurgitation characteristics.
We designed a Transformer-based deep learning model for fully automatic Doppler video detection of MR. This framework can automatically detect MR and classify central or eccentric regurgitation. The algorithm was trained, validated, and tested using retrospectively selected studies. A prospective dataset of 217 patients was used as an independent test database.
(a) The model demonstrated exceptional diagnostic accuracy for MR, achieving accuracy rates of 0.94(95%CI: 0.90-0.97) and 0.92(95%CI: 0.89-0.96) in retrospective and prospective test sets, respectively, with area under the curve (AUC) values of 0.98 (95%CI: 0.96-1.0) and 0.97 (95%CI: 0.95-0.99). Its performance aligned with that of middle seniority physicians. (b) The model achieved an accuracy of 0.93 in identifying central and eccentric MR across both retrospective and prospective test sets, AUC values of 0.96 (95%CI: 0.89-0.99) and 0.95 (95%CI: 0.89-0.98), respectively. (c) On the tasks of MR, the diagnostic performance of the model in multiple views was better than that of the parasternal long-axis view (PLAX) alone or the apical four-chamber view (A4C) in both retrospective and prospective datasets.
The Transformer-based algorithm proposed in this study can automate and improve the efficiency of clinical workflows, screen multi-view for the presence of MR, and assist in classifying regurgitation characteristics.
Authors
Zhong Zhong, Zhang Zhang, Sheng Sheng, Guo Guo, Yang Yang, Huang Huang, Zhao Zhao, Zeng Zeng, Chen Chen, Ni Ni, Xu Xu, Xue Xue, Liu Liu
View on Pubmed