A Comparative Analysis of Deep Convolutional Networks for Automated Diagnosis of Retinal Detachment in Dogs.

ResNet50V2 shows strong promise for automated retinal detachment screening in dogs.

This multicenter retrospective study evaluated three deep convolutional neural networks (ResNet50V2, VGG16, EfficientNetB0) for automated detection of retinal detachment in canine fundus photographs. Using 2000 color fundus images (793 with retinal detachment, 1207 normal) from 275 dogs collected between 2020-2025, researchers applied transfer learning with standardized preprocessing and real-time augmentation. The dataset was split at the patient level (80% training, 20% validation) to ensure independent validation. ResNet50V2 demonstrated superior performance with 89.09% accuracy and 0.9194 AUC, followed by EfficientNetB0 (81.82% accuracy, 0.8831 AUC). VGG16 showed poor reliability (61.82% accuracy, 0.6868 AUC) due to high false-positive rates. Gradient-weighted class activation mapping confirmed the best model appropriately focused on retinal detachment regions. While ResNet50V2 shows promising potential as a screening support tool for canine retinal detachment, the authors recommend prospective external validation across multiple devices and clinical settings before implementing routine clinical use.

This summary was distilled by AI and may occasionally misinterpret data. Confirm critical details with the primary literature before clinical application.