Fig. 1: SLIVER-net performance.
From: Automated identification of clinical features from sparsely annotated 3-dimensional medical imaging
SLIVER-net (dark blue) was compared with a 3D-CNN backbone approach (light blue) and 2D CNN (gray). SLIVER-net significantly outperformed both the 3D CNN and 2D CNN in identifying each biomarker in terms of area under the ROC (AUROC) and area under the precision-recall curve (precision-recall AUC). Top: Precision-recall AUC for each biomarker. Bottom: ROC AUC for each biomarker. Horizontal bars indicate a significant difference in performance between the two models. Error bars represent 95% confidence interval (CI) calculated using a bootstrapping procedure.
