RESUMO
PURPOSE: To develop an automated deep learning system for detecting the presence and location of disc hemorrhages in optic disc photographs. DESIGN: Development and testing of a deep learning algorithm. METHODS: Optic disc photos (597 images with at least 1 disc hemorrhage and 1075 images without any disc hemorrhage from 1562 eyes) from 5 institutions were classified by expert graders based on the presence or absence of disc hemorrhage. The images were split into training (n = 1340), validation (n = 167), and test (n = 165) datasets. Two state-of-the-art deep learning algorithms based on either object-level detection or image-level classification were trained on the dataset. These models were compared to one another and against 2 independent glaucoma specialists. We evaluated model performance by the area under the receiver operating characteristic curve (AUC). AUCs were compared with the Hanley-McNeil method. RESULTS: The object detection model achieved an AUC of 0.936 (95% CI = 0.857-0.964) across all held-out images (n = 165 photographs), which was significantly superior to the image classification model (AUC = 0.845, 95% CI = 0.740-0.912; P = .006). At an operating point selected for high specificity, the model achieved a specificity of 94.3% and a sensitivity of 70.0%, which was statistically indistinguishable from an expert clinician (P = .7). At an operating point selected for high sensitivity, the model achieves a sensitivity of 96.7% and a specificity of 73.3%. CONCLUSIONS: An autonomous object detection model is superior to an image classification model for detecting disc hemorrhages, and performed comparably to 2 clinicians.