EyeSmart A Deep Learning and Random Forest Approach for Multi-Disease Retinal Analysis

Six eye diseases; Diabetic Retinopathy (DR), Glaucoma, Age-related Macular Degeneration (AMD), Cataracts, Retinal Detachment, and Keratitis are featured in this work with an automated procedure of detection and classification. Consequently, this education aims at enhancing the work of automated detection systems since manual screening proves complicated, requires professionals’ attention, and costs a lot of money. The use of the proposed model integrates a Random Forest classifier with a Convolutional Neural Network (CNN) based on the strength of the two methods. In the model development, a proper data preparation process was applied where 3,645 retinal pictures from different sources were used. In order to create the training set, images were rescized and normalised and further image enhancing was also done. More detailed information from the pictures of the retina should be revealed by the CNN consisting of six convolutional layers and two max-pooling layers.Configured with 32, 64, 128, 256, 512, and 1024 filters correspondingly, the convolutional layers faithfully caught complex patterns in the photos. After flattening the acquired features, they were sent to a Random Forest classifier with 15 trees and an 8 maximum depth for ultimate classification. With an overall accuracy of 88.44%, the model’s performance was assessed on demanding datasets including APTOS-2019 and IDRiD using high precision, recall, and F1-scores across all six eye illnesses. F1-scores for DR, Glaucoma, AMD, Cataracts, Retinal Detachment, and Keratitis were particularly 87.74% for DR, $\mathbf{8 9. 6 6 \%}$ for Glaucoma, $\mathbf{8 6. 6 7 \%}$ for AMD, $\mathbf{8 5. 0 9 \%}$ for Cataracts, 89.64% for Retinal Detachment, and 91.17% for Keratitis. These results show the promise of the suggested hybrid model to greatly enhance automated screening procedures, therefore lowering reliance on human analysis and increasing early diagnosis and treatment results for many eye disorders. Future studies will concentrate on improving the model even further and verifying its efficacy in several clinical environments.