Open Internet by MindsNet

Improving Consistency in Diabetic Retinopathy Classification Models

A 5-class Diabetic Retinopathy classifier trained on the APTOS 2019 dataset shows inconsistent predictions across classes, with high confidence in incorrect predictions. The model struggles with class confusion, particularly in distinguishing between Moderate, Severe, and Proliferative DR. The author seeks advice on addressing this issue, which may be due to domain shift, model limitations, preprocessing issues, or class imbalance.

Computing & Technology, Computer Science, Artificial Intelligence