A Reliable Diabetic Retinopathy Grading via Transfer Learning with Quadratic Weighted Kappa Metric

Diabetic Retinopathy (DR), the most prevalent eye condition in diabetic patients, can be brought on by diabetes. It may result in reduced eyesight or possibly complete blindness. Early detection can help prevent eye damage, but it can be challenging because symptoms might not show up right away. To grade diabetic retinopathy and identify all stages of DR, many predictive models based on machine learning and deep learning have been proposed. However, the majority of the diabetic retinopathy grading models now in use cannot identify all stages. For DR grading, the most used metrics, including as accuracy, f1-score, precision, recall, and AUC-ROC score, are unreliable. Since they do not consider the level of disagreement between target and predicted labels and the ordinality of target labels, which helps classify all DR stages. To address these issues, we developed the customized EfficientNet b3. The Quadratic Weighted Kappa (QWK) was chosen as the primary evaluation metric since the distinct stages of diabetic retinopathy are ordinal. The degree of discrepancy between the actual and expected labels is also taken into consideration. We were able to achieve a higher QWK score of 0.87 by recognizing all the DR stages after aggregating the predictions from the efficientnet-b3 model, which was trained for 30 and 60 epochs, respectively. This result is state-of-the-art. DR grading must identify each stage in order to provide patients with the greatest care available. Our Confusion matrix of ensembled efficientnetb3 shows, which is state-of-the-art, that all DR stages are identified with higher probabilities.