Automatic classification of optical coherence tomography images into normal/ARMD class using deep transfer learning and arithmetic optimizer

Eye is a vital sensory organ in human physiology and abnormality in eye will lead to various vision related illnesses. Appropriate diagnosis and treatment is necessary to reduce the impact of eye abnormality. The objective of this research is to develop an automated scheme based on deep-learning (DL) for detecting Age-Related Macular Degeneration (ARMD). Further, this work considered the Arithmetic Optimizer (AO). Based features selection and fusion to achieve significant result without over fitting issue. Various phases in the proposed DL-scheme includes; (i) labelled images collection, and resizing it to required dimension, (ii) deep-features extraction using DL-model and implementing softmax-based binary classification, (iii) identification of best two DL-model features and optimizing its value with AO and generating the fused-features using serial features concatenation, and (iv) verification of the performance of the developed system using 5-fold cross validation process. This research considered the pre-trained DL-models for verifying the performance of the developed tool using 3000 images per class (Normal/ARMD). The DL-system’s performance is verified using conventional-features, AO optimized-features, and fused-features and the experimental outcome confirms that the DL-model based approach help to achieve system help to achieve 99.8333% accuracy with K-nearest neighbour classifier on the chosen Optical Coherence Tomography (OCT) database.