Abstract
Diabetic Retinopathy (DR) is one of the common complications of diabetic patients, which can
lead to vision loss or even blindness. Early detection and timely treatment are crucial for
preventing vision loss. However, in real-world applications, DR classification tasks often face the
challenge of imbalanced data distributions, where one class significantly outnumbers the others.
Therefore, this research proposed on a novel ensemble learning model which combines ResNet50
and Dual attention method for Diabetic Retinopathy classification task trained on the pre
processed APTOS 2019 Blindness Detection datasets that were downloaded from Kaggle. The
model outperformed the pre-training model with an Average Precision (AP) of 1 in the No DR
class and an Area Under the Curve (AUC) of 0.97 in the Advanced class, 0.93 in the Mild class, 1
in the No DR class, and 0.96 in the Proliferative DR class, respectively. Besides, this paper utilizes
local interpretable Model Interpretation (LIME) as the interpretable AI tool. The integrated model
that combines ResNet50 with the dual attention strategy proposed in this research performs
admirably in the diabetic retinopathy classification test. It has a high potential for medical
auxiliary diagnosis and exhibits strong discriminative ability and good interpretability,
particularly under imbalanced data.