Decoding Machine Learning Algorithms for DR Detection

Diabetic retinopathy is a serious eye disease which can lead to vision defects in diabetic patients. Early detection is important for preventing vision loss. Automating the detection process makes it less laborious and helps in achieving more accurate results. Therefore, this research aims to study various machine learning methods and develop an effective model for diabetic retinopathy detection. The proposed model operates through four stages: pre-processing, feature extraction, feature optimization, and classification. Image processing techniques are thoroughly utilized to enhance the fundus images. The textual and spatial features are then extracted and most relevant features are selected using particle swarm optimization. These features are then fed into machine learning classifiers for disease diagnosis. The model is trained and validated on the MESSIDOR I and MESSIDOR II datasets. The model achieves an accuracy (85.9%), F1 score (0.8585), Sensitivity (0.8049) and Specificity (0.8958) on MESSIDOR II data with Support Vector Machine (SVM) classifier. The results obtained show the effectiveness of the work in early detection.