Hybrid Image Features Supported Normal/Abnormal Retinal Optical Coherence Tomography Image Classification

Background Eye is one of the primary sensory organs, responsible to provide the necessary visual information to the brain. Any abnormality in eye cause mild to severe vision issues and hence appropriate detection and treatment is preferred. Clinical level examination of the eye is commonly performed using image examination procedures and this work considered the Optical Coherence Tomography (OCT) for the study. In the present study, the proposed system aims to develop a tool to classify the OCT-images into normal and abnormal class using the traditional deep-learning (DL) and machine-learning (ML) features. Methods This tool consists the following stages; OCT-image collection and resizing based on the chosen DL scheme, image features extraction using DL and ML methods, DL feature reduction using 50% dropout and serially fusing the deep- and machine-features to generate hybrid-features, and binary classification using 5-fold cross-validation. Results The developed tool’s merit is verified using; (i) deep-features, (ii) machine-features, and (iii) hybrid-features. The merit of the developed scheme is verified using different binary classifiers and the overall quality metric is considered to verify the tool’s performance. Conclusions The study confirms that the hybrid-features based detection provides 97.6% accuracy, when Random Forest classifier is employed, which verifies the tool’s merit on the chosen OCT database. In the future, this tool can be considered to detect the common AEAs, like AMD and the DME.