ENHANCED DIABETIC EYE DISEASE CLASSIFICATION USING INCEPTIONV3 OPTIMIZED WITH THE GENETIC ALGORITHM AND NOVEL EYENET WEIGHTS

Purpose:

To improve the accuracy and efficiency of diabetic eye disease classification by developing a deep learning model optimized using the Genetic Algorithm (GA) and fine-tuned with novel EYENET weights, specifically designed for ocular surface images.

Setting:

The study utilized a newly curated Di-EYENET dataset comprising 3000 ocular surface images collected from 1000 healthy (non-diabetic) individuals, 1000 diabetic Type-1 patients, and 1000 diabetic Type-2 patients. The dataset was validated over an eight-month period by medical professionals to ensure its accuracy and reliability.

Method:

We employed the InceptionV3 deep learning architecture, optimized using the Genetic Algorithm (GA) to fine-tune the model’s hyperparameters. Additionally, the model was trained using EYENET weights, specifically designed for diabetic eye disease classification, as opposed to traditional ImageNet weights. The GA optimization was applied to enhance feature extraction and model performance across different categories of diabetic eye disease.

Results:

Our GA-optimized InceptionV3 model, fine-tuned with EYENET weights, demonstrated a 7% increase in accuracy compared to models trained with traditional ImageNet weights. The model showed enhanced classification accuracy in distinguishing between diabetic Type-1, diabetic Type-2, and non-diabetic ocular surface images. The results indicate significant improvements in diagnostic precision, particularly in diabetic retinopathy detection.

This study presents a robust and efficient deep learning model for diabetic eye disease classification. By combining the Genetic Algorithm with a domain-specific dataset and tailored EYENET weights, our approach offers a significant advancement in diagnostic accuracy, making it a promising tool for automated diabetic eye disease detection in clinical settings.