Title : Automated interpretation of congenital heart disease from multi-view echocardiograms
Abstract:
Congenital heart disease (CHD) is the most common birth defect and the leading cause of neonate death in China. Clinical diagnosis can be based on the selected 2D key-frames from five views. Limited by the availability of multi-view data, most methods have to rely on the insufficient single view analysis. This study proposes to automatically analyze the multi-view transthoracic echocardiogram (TTE) echocardiograms with a practical end-to-end framework and build an automatic and interpretable assistant for the TTE-based assessment of atrial septal defect (ASD).
Firstly, we collect the five-view echocardiograms video records of CHD and healthy subjects with both disease labels and standard-view key-frame labels. Depthwise separable convolutionbased multi-channel networks are adopted to largely reduce the network parameters. We also approach the imbalanced class problem by augmenting the positive training samples. Our 2D key-frame model can diagnose CHD or negative samples with an accuracy of 95.4%, and in negative, VSD or ASD classification with an accuracy of 92.3%. To further alleviate the work of key-frame selection in real-world implementation, we propose an adaptive soft attention scheme to directly explore the raw video data. Our video-based model can diagnose with an accuracy of 93.9% (binary classification), and 92.1% (3-class classification) in a collected 2D video testing set, which does not need key-frame selection and view annotation in testing.
Secondly, we developed a novel deep keypoint stadiometry (DKS) model, which learns to precisely localize the keypoints, i.e., the endpoints of defects, and followed by the absolute distance measurement with the scale. The closure plan and the size of the ASD occluder for transcatheter closure are derived based on the explicit clinical decision rules. The accuracy of closure classification using DKS (0.9425±0.0052) outperforms “black-box” model (0.7646±0.0068; p<0.0001) for within-center evaluation. The results in cross-center cases or using the quadratic weighted kappa as an evaluation metric are consistent. The fine-grained keypoint label provides more explicit supervision for network training. While DKS can be fully automated, clinicians can intervene and edit at different steps of the process as well.
The presented model has high diagnostic rates for VSD and ASD that can be potentially applied to the clinical practice in the future. The short-term automated machine learning process can partially replace and promote the long-term professional training of primary doctors, improving the primary diagnosis rate of CHD in China, and laying the foundation for early diagnosis and timely treatment of children with CHD.
