Protecting and Mitigating Against Cyber Threats

25Closing the Security Gap: Towards Robust and Explainable AI for Diabetic Retinopathy

R. S. M. Lakshmi Patibandla^*

Department of CSE, Koneru Lakshmaiah Education Foundation, AP, India

Abstract

Reducing blindness from diseases such as Diabetes which causes Diabetic retinopathy is a testimony of how the application of the two models of ML and AI in early diagnosis and treatment. Thus, despite the relatively futuristic concept of diabetic retinopathy screening and application of AI for image analysis of the retinal photographs, there are some limitations, the primary ones being invulnerability points and neutrality. In response to this major deficiency, this chapter proposes a framework to enhance and use trustworthy AI for DR diagnosis.

The researcher also describes the risks that are associated with the acquisition of the DR diagnosis via AI; they are data poisoning, adversarial attacks, and privacy invasion. The chapter also describes how such risks can be avoided and, once again, appeals to the proper definition of the model, the determination of explainable AI, and the non-sharing of unsuitable data. Therefore, due to such considerations, this framework offers a chance of building the safe and ethical cooperative relations with the AI for the improvement of the DR diagnosis.

Keywords: Medical conditions, diabetic retinopathy, AI, ML in security, XAI, data poisoning, adversarial attack, privacy

25.1 Introduction

A common disease Diabetic retinopathy (DR) affects the number ...