Explainable Artificial Intelligence for Event and Anomaly Detection in Healthcare Monitoring Systems

The growing uses of artificial intelligence (AI) in the form of beside vital-sign monitors, wearable sensors or Internet of Medical Things (IoMT) devices allows early identification of critical events and various anomalies including arrhythmias, sepsis onset, falls, or physiological deterioration. The underlying deep learning and ensemble models, however, are typically opaque, undermining confidence in their clinical recommendations, regulatory approval, and patient safety in the case of an unsubstantiated event detection. Explainable Artificial Intelligence (XAI) aims to address this challenge by providing justifications and explanations accompanying the model's predictions that are understandable to a human. The paper provides an overview of the state-of-theart in XAI for events and anomalies detection in healthcare monitoring and discusses opportunities and barriers to adopting XAI methods in the application domain. We summarize the main explanation approaches, provide an overview of their use in the context of electrocardiogram (ECG) analysis, IoMT security, behavioural healthcare monitoring and ICU deterioration prediction, and articulate a general conceptual framework. The paper concludes with a discussion of open challenges and future research direction.