Normal/Cataract Detection in Fundus Image Using Individual and Fused ResNet Features

The visual sensory information collected by the eyes is crucial for accurate perception and decision-making in the brain. Any ocular ailment will disrupt this process, perhaps resulting in mild to severe vision-related complications. Ocular ailments are primarily attributable to disease or senescence. Age-related eye illness is a prevalent concern that necessitates prompt detection and intervention. Cataract is a prevalent age-related ocular condition that results in mild to severe visual impairment and necessitates a small surgical intervention for correction. The image-guided identification of cataracts is a clinical practice, and this research intends to present a Deep Learning (DL) method to categorize Retinal Fundus Images (RFI) as normal or cataract. The proposed scheme comprises several phases: (i) image acquisition and resizing to 224×224 pixels, (ii) feature extraction utilizing a DL- model, (iii) optimal model selection, feature reduction with 50% dropout, and concatenation of serial features, and (iv) classification accompanied by 3-fold cross-validation to validate performance. This study evaluates the efficacy of the suggested DL-tool utilizing both traditional and fused-features. The experimental results of this work demonstrate that the fused-features technique achieves > 98% accuracy when applied to SM-based categorization.