A Comparative Study and Systematic Analysis of XAI Models and their Applications in Healthcare

Abstract

Artificial intelligence technologies such as machine learning and deep learning employ techniques to anticipate results more effectively without human involvement. Since AI models are viewed as opaque models, their application in healthcare is still restricted. Explainable artificial intelligence (XAI) has been designed to increase the use of artificial intelligence (AI) algorithms in the healthcare sector by increasing trust in the model's predictions and explaining how they are developed. The aim of this article is to critically review, compare, and summarize existing research and to find new research possibilities of XAI for applications in healthcare. This study is conducted by finding articles related to XAI in biological and healthcare domains from the PubMed, Science Direct, and Web of Science databases using the PRISMA method. A comparative study of the state-of-the-art XAI techniques to evaluate its applications in healthcare has also been done using an experimental demonstration on the Diabetes dataset. XAI techniques, namely LIME, SHAP, PDP, and decision tree, were used to explain how various input attributes contributed to the outcome of the model. This study found that the explanations provided by these models are not easily understandable for different users of the model, like doctors and patients, and need expertise. This study found that the potential of XAI in the medical domain is high as it increases trust in the AI model. This survey will motivate the researchers to build more XAI techniques that provide user-friendly explanations, especially for the less explored areas of medical data, such as biomedical signals and biomedical text.