Extended Data Table 8 Multi-view DNN performance stratified by echo machine manufacturer
From: Multiview deep learning improves detection of major cardiac conditions from echocardiography
Left/Right Ventricular Abnormality (value (95%CI)) | AUC | Sensitivity | Specificity | PPV | NPV | F1 |
|---|---|---|---|---|---|---|
All data (N=6279) | 0.907 (0.900–0.914) | 0.810 (0.794–0.827) | 0.840 (0.832–0.849) | 0.609 (0.591–0.628) | 0.935 (0.929–0.941) | 0.695 (0.679–0.710) |
Phillips (N=4,720) | 0.906 (0.898–0.915) | 0.801 (0.781–0.821) | 0.853 (0.844–0.862) | 0.631 (0.611–0.651) | 0.932 (0.925–0.939) | 0.706 (0.689–0.722) |
GE (N=1,022) | 0.913 (0.896–0.931) | 0.810 (0.768–0.850) | 0.845 (0.825–0.866) | 0.598 (0.553–0.645) | 0.940 (0.925–0.954) | 0.688 (0.651–0.724) |
Acuson (N=508) | 0.913 (0.888–0.938) | 0.743 (0.673–0.813) | 0.901 (0.876–0.924) | 0.661 (0.592–0.729) | 0.931 (0.909–0.952) | 0.700 (0.640–0.755) |
Diastolic Dysfunction | ||||||
All data (N=2263) | 0.836 (0.821–0.851) | 0.736 (0.718–0.755) | 0.774 (0.747–0.801) | 0.878 (0.864–0.893) | 0.570 (0.544–0.596) | 0.801 (0.787–0.815) |
Phillips (N=1723) | 0.824 (0.806–0.843) | 0.757 (0.736–0.777) | 0.726 (0.695–0.757) | 0.868 (0.850–0.885) | 0.557 (0.523–0.590) | 0.808 (0.794–0.824) |
GE (N=394) | 0.907 (0.880–0.930) | 0.659 (0.612–0.707) | 0.953 (0.924–0.979) | 0.959 (0.933–0.982) | 0.627 (0.572–0.677) | 0.781 (0.746–0.816) |
Valve Regurgitation | ||||||
All data (N=4148) | 0.904 (0.892–0.915) | 0.826 (0.796–0.854) | 0.829 (0.818–0.839) | 0.395 (0.369–0.421) | 0.972 (0.967–0.977) | 0.534 (0.509–0.560) |
Phillips (N=2721) | 0.907 (0.894–0.921) | 0.825 (0.792–0.860) | 0.837 (0.825–0.849) | 0.408 (0.379–0.439) | 0.972 (0.967–0.978) | 0.546 (0.517–0.576) |
GE (N=1069) | 0.915 (0.897–0.933) | 0.814 (0.770–0.855) | 0.864 (0.845–0.885) | 0.630 (0.581–0.678) | 0.942 (0.928–0.956) | 0.710 (0.669–0.749) |
Acuson (N=271) | 0.920 (0.892–0.948) | 0.852 (0.772–0.925) | 0.820 (0.778–0.863) | 0.541 (0.453–0.630) | 0.957 (0.930–0.979) | 0.662 (0.583–0.731) |
