XAI Unveiled: Revealing the Potential of Explainable AI in Medicine: A Systematic Review

Abstract

Nowadays, artificial intelligence in medicine plays a leading role. This necessitates the need to ensure that artificial intelligence systems are not only high-performing but also comprehensible to all stakeholders involved, including doctors, patients, healthcare providers, etc. As a result, the explainability of artificial intelligence systems has become a widely discussed subject in recent times, leading to the publication of numerous approaches and solutions. In this paper, we aimed to provide a systematic review of these approaches in order to analyze their role in making artificial intelligence interpretable for everyone. The conducted review was carried out in accordance with the PRISMA statement. We conducted a BIAS analysis, identifying 87 scientific papers from those retrieved as having a low risk of BIAS. Subsequently, we defined a classification framework based on the classification taxonomy and applied it to analyze these papers. The results show that, although most AI approaches in medicine currently incorporate explainability methods, the evaluation of these systems is not always performed. When evaluation does occur, it is most often focused on improving the system itself rather than assessing users’ perception of the system’s effectiveness. To address these limitations, we propose a framework for evaluating explainability approaches in medicine, aimed at guiding developers in designing effective human-centered methods.

Keywords

• Explainability
• artificial intelligence
• interpretability
• medicine
• explainable artificial intelligence
• interpretable artificial intelligence