AN INTERPRETABLE MACHINE LEARNING FRAMEWORK FOR THE DETECTION OF DIABETIC RETINOPATHY

Abstract

Continuous movements in Artificial Intelligence (AI) and the addition of computational assets and abilities have set out the opportunity to cultivate Deep Learning (DL) applications for the detection and classification of diabetic retinopathy (DR). It offers most advantageous results. But it is tracked down that the model's general precision is not adequate all alone to permit clinicians to settle on a machine learning (ML) model. Clinicians see reasonableness as a method for legitimizing their clinical dynamic with regards to a model's choice.
Subsequently, there is a need of hour to plan ML models working with the justification measure. This article proposes a model agnostic method on the top of ML model to provide explainabliity and interpretability for underlying model for the detection and classification of diabetic retinopathy (DR). It has a great advantage of flexibility by portioning the explainability from ML models. This framework will provide best results to enhance model interpretability, making clinical decisions more robust, bridging the gap between ML solution & human explanations and make better acceptance of ML / AI in sensitive & critical domains where value of human life is of an enormous concern such as healthcare