Extended Data Table 2 Performance of multi-view DNNs in disease strata in the UCSF test dataset
From: Multiview deep learning improves detection of major cardiac conditions from echocardiography
Ventricular Abnormality Type (value (95%CI)) | AUC | Sensitivity | Specificity | F1 |
|---|---|---|---|---|
Abnormal LV Size (>86 ml/m2 for males, >70 ml/m2 for females) (N=4888) | 0.965 (0.946–0.983) | 0.929 (0.880–0.975) | 0.840 (0.831–0.848) | 0.170 (0.143–0.198) |
Abnormal LV Function (LVEF <50%) (N=5692) | 0.949 (0.943–0.955) | 0.901 (0.885–0.918) | 0.840 (0.831–0.848) | 0.652 (0.633–0.67) |
Abnormal RV Size (≥moderately enlarged) (N=5262) | 0.903 (0.890–0.915) | 0.806 (0.774–0.837) | 0.840 (0.831–0.849) | 0.463 (0.438–0.489) |
Abnormal RV Function (≥moderately decreased) (N=5125) | 0.950 (0.939–0.959) | 0.891 (0.861–0.92) | 0.840 (0.831–0.849) | 0.417 (0.389–0.441) |
Diastolic Dysfunction by grade (value (95%CI)) | AUC | Sensitivity | Specificity | F1 |
Grade 1–2 (N=1765) | 0.836 (0.821–0.853) | 0.734 (0.71–0.756) | 0.777 (0.751–0.805) | 0.784 (0.767–0.8) |
Grade 3 (N=1022) | 0.836 (0.817–0.856) | 0.736 (0.696–0.776) | 0.777 (0.753–0.804) | 0.677 (0.644–0.708) |
Grade 4 (N=1088) | 0.843 (0.824–0.862) | 0.742 (0.707–0.778) | 0.777 (0.751–0.803) | 0.704 (0.674–0.732) |
Regurgitation by valve (value (95%CI)) | AUC | Sensitivity | Specificity | F1 |
Aortic Valve (N=3719) | 0.891 (0.863–0.918) | 0.769 (0.677–0.855) | 0.829 (0.818–0.839) | 0.135 (0.106–0.162) |
Mitral Valve (N=3882) | 0.943 (0.932–0.953) | 0.904 (0.868–0.933) | 0.829 (0.819–0.839) | 0.389 (0.361–0.423) |
Tricuspid Valve (N=3916) | 0.885 (0.868–0.902) | 0.790 (0.747–0.831) | 0.829 (0.819–0.839) | 0.378 (0.346–0.408) |
