Healthy/Glaucoma Fundus Retinal Image Classification using Butterfly Algorithm Optimized ResNet-Features

Hospitals frequently use medical image supported disease diagnosis to identify diseases and evaluates their severity. Automation in image examination is normally preferred to reduce the diagnostic burden. Developing a Deep-Learning Tool (DLT) to analyze the fundus-photography (FP) labeled as healthy/glaucoma class is the aim of this research. Different phases in this DLT involves the following; collection of FP database from image repository and resizing it to 224x224x3 pixels, feature extraction using ResNet (RN) model, feature reduction using the Brownian-Butterfly Algorithm (BBA), serial concatenation of features to obtain fused-features (FF), and executing the binary classification using 3-fold cross validation. This work utilized a two-fold training in order to improve detection accuracy with the FP data. This investigation is executed using individual and FF to verify the DLT's performance. The results of this research validates that, the K-Nearest Neighbor (KNN) assisted classification provides 100% accuracy when FF based investigation is executed.