Fig. 3: Confusion matrices for SLIVER-net and the three retinal specialist annotators.
From: Automated identification of clinical features from sparsely annotated 3-dimensional medical imaging
100 of the 390 test set patients were selected for comparison with clinician performance. The remaining 290 patients were used to compute the SLIVER-net threshold, which was selected to match the mean sensitivity of the annotators. For Subretinal Drusenoid Deposits and Intraretinal HRF, SLIVER-net displays a similar sensitivity to clinicians while operating at fewer false positives.
