Publication

Abstract : Age-related Macular Degeneration AMD, Cataract, Diabetic Retinopathy DR and Glaucoma are the four most common ocular conditions that affect a person's vision. Early detection in the asymptomatic stages can alleviate vision loss or slow down the progression of these diseases. However, manual diagnosis is a costly and tedious process, especially in mass screening applications. Computer aided diagnosis CAD systems serve as an aid to ophthalmologists in efficient diagnosis of ocular diseases. In particular, a generic CAD framework that detects multiple ocular diseases could be immensely beneficial. In the proposed work, a single framework for detection of the above-mentioned ocular diseases has been explored. Specifically, a pool of non-linear handcrafted features are extracted from fundus images, followed by feature selection using a hybrid optimization algorithm, where features are selected using JAYA algorithm JA first, followed sequentially by the Harris hawks optimization HHO algorithm. The selected features are used to train an extreme gradient boosting XGB model for disease classification. Unlike existing systems that restrict non-linear features to single ocular disease detection, the proposed system is the first of its kind for detection of the above-specified multiple ocular diseases in a generic framework, yielding 93 % accuracy, 91.3 % sensitivity, 96.4 % specificity, 90.4 % precision and 90.8 % F1 score. Further, in this study, Shapley additive explanations SHAP analysis is employed to gain insight on the impact of the non-linear features on the model's prediction capability. This work clearly demonstrates the importance of explainability that opens the ‘black box’ nature of machine learning ML model and clearly unveils the relationships among the features and the diagnosis. Also, the explainable ML model improves transparency of the model's decision-making process. The proposed algorithm can efficiently assist physicians in diagnosing the ocular diseases using fundus images in clinical practice, and avoids the subjective difference that comes with manual assessment.