Physiological signal analysis using explainable artificial intelligence: A systematic review

With the continuous development of wearable sensors, it has become increasingly convenient to collect various physiological signals from the human body. The combination of Artificial Intelligence (AI) technology and various physiological signals has significantly improved people's awareness of their psychological and physiological states, thus promoting substantial progress in the medical and health industries. However, most current research on physiological signal modeling does not consider the issue of interpretability, which poses a significant challenge for clinical diagnosis and treatment support. Interpretability refers to the explanation of the internal workings of AI models when generating decision results and is regarded as an important foundation for understanding model operations. Despite substantial progress made in this field in recent years, there remains a lack of systematic discussion regarding interpretable AI in physiological signal modeling, resulting in researchers having difficulty comprehensively grasping the latest developments and emerging trends in the field. Therefore, this paper provides a systematic review of interpretable AI technologies in the domain of physiological signals. Based on the scope of interpretability, these technologies are divided into two categories: global and local interpretability, and we conduct an in-depth analysis and comparison of these two types of technologies. Subsequently, we explore the potential applications of interpretable physiological signal modeling in areas such as medicine and healthcare. Finally, we summarize the key challenges of interpretable AI in the context of physiological signals and discuss future research directions. This study aims to provide researchers with a systematic framework to better understand and apply interpretable AI technologies and lay the foundation for future research.