Explainable Artificial Intelligence Evaluation for Healthcare: A Literature Review : Emerging Trends in Digital Health, Summer Term 2023

Background: Explainable Artificial Intelligence (XAI) is increasingly important in healthcare, where transparency and trust are crucial. Traditional AI's "black box" nature often interferes with acceptance among medical professionals and patients, despite its potential to improve diagnostic and treatment processes. XAI approaches proposed in the literature have not yet been evaluated in practice by health professionals due to lack of time and conviction. Without XAI, doctors would have to resort to manual and traditional data processing methods, which are less precise and time-consuming. The consequences can be delayed and incorrect diagnoses, overloading of staff, and avoidable serious illnesses.
Objective: XAI methods depend on their context, especially their target group. At present, the research community lacks appropriate and mature evaluation approaches that convincingly assess XAI. This literature review aims to provide a comprehensive overview of existing evaluation methods in healthcare and to identify key issues and challenges in this interdisciplinary field.
Methods: Using Braun and Clarke's thematic analysis, literature from Scopus was systematically reviewed, focusing on peer-reviewed journals. ... mehrAs relevant identified literature was coded, and themes were derived to understand the current state of XAI evaluation methods in healthcare.
Results: Using a collection of relevant literature and a selection of codes, eight key themes were identified: XAI, experiment circumstances, stakeholders, evaluation criteria, evaluation methods, evaluation metrics, ethical and legal implications, and challenges.
Conclusion: This work underscores the necessity for standardized approaches and definitions in XAI evaluation to enhance comparability and reliability. Future research should focus on interdisciplinary collaboration and developing comprehensive, context-specific evaluation frameworks to advance XAI implementation in healthcare.