Wolf Social Leader optimized BiLSTM for diabetic retinopathy detection and grading

Diabetic retinopathy (DR) is a common complication of diabetes mellitus and a major cause of vision loss worldwide. Early detection and grading of DR are crucial for timely intervention and effective treatment. The latest advancement in machine learning is seen as a solution for early DR detection and grading of DR by processing fundus images. However, many existing models have limitations such as overfitting and high computational complexity. In addition, several models exhibit insufficient robustness and poor generalization when dealing with complex and heterogeneous fundus images. They fail to capture subtle texture-based features. To address these challenges, this study proposes a novel Wolf Social Leader algorithm-enabled Bi-directional Long Short-Term Memory (WS-BiLSTM) model for accurate DR detection and grading. The proposed approach integrates weighted shape-based texture patterns and statistical feature extraction to selectively emphasize subtle texture-based features. The model disregard irrelevant information as a means to mitigate the overfitting problem. The model uses ResNet-101 to extract informative deep representations from fundus images, which enhances classification accuracy. The Wolf Social Leader (WSL) algorithm is incorporated to optimize the model hyperparameters, which results in reduced computational cost. Experimental results demonstrate that our proposed WS-BiLSTM model achieves 96.32% accuracy, 97.21% sensitivity, and 95.42% specificity. Thereby, outperforming several existing models. These results demonstrate the potential of our proposed model to support clinicians by enabling more accurate and reliable decision-making, thereby contributing to earlier diagnosis and grading of DR.