Extended Data Table 4 Performance of multi-view DNNs at sensitivity- and specificity-optimized thresholds in the UCSF test dataset

From: Multiview deep learning improves detection of major cardiac conditions from echocardiography

DNN class

Sensitivity-Optimized

Specificity-Optimized

Fixed Sensitivity

Specificity

Sensitivity

Fixed Specificity

Left/Right Ventricular Abnormality cohort, test set (N=6279)

LVRV Multi-view DNN

(value (95%CI))

0.800 (0.783–0.817)

0.851 (0.843–0.859)

0.837 (0.821–0.853)

0.800 (0.790–0.809)

Diastolic dysfunction cohort, test set (N=2214)

Diastolic Dysfunction Multi-view DNN

(value (95%CI))

0.800 (0.782–0.817)

0.721 (0.692–0.749)

0.691 (0.671–0.710)

0.800 (0.773–0.825)

Valve regurgitation cohort, test set (N=4148)

Valve Regurgitation Multi-view DNN

(value (95%CI))

0.800 (0.770–0.832)

0.845 (0.836–0.855)

0.850 (0.822–0.877)

0.800 (0.789–0.810)

  1. The threshold of each trained Multi-view DNN was selected to fix sensitivity or specificity to 0.800 in the test dataset as indicated. (DNN) Deep Neural Network, (CI) Confidence Interval. All values are point estimate (95% confidence interval, derived via bootstrap).
Back to article page