Exploring Explainable AI Techniques for Text Classification in Healthcare: A Scoping Review.

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Machine Learning

Background:

• Text classification is crucial for organizing medical data using machine learning (ML) and deep learning (DL).
• The 'black box' nature of AI in healthcare raises concerns about model interpretability.
• Explainable AI (XAI) offers methods to understand AI decision-making processes.

Purpose of the Study:

• To conduct a scoping review of XAI techniques applied to medical text classification.
• To identify and categorize different types of XAI methods used in this domain.
• To assess the current state of evaluation and user adoption of XAI in healthcare.

Main Methods:

• Systematic literature search for XAI applications in medical text classification.
• Categorization of identified XAI techniques into model-specific and model-agnostic approaches.
• Analysis of studies reporting on the effectiveness and usability of XAI.

Main Results:

• Two primary categories of XAI methods were identified: model-specific and model-agnostic.
• While developers show some positive reception, formal evaluations with medical end-users are scarce.
• Limited studies exist on the practical impact and acceptance of XAI by clinicians.

Conclusions:

• XAI techniques show promise for increasing transparency in AI-driven medical text classification.
• There is a significant need for more rigorous evaluations involving healthcare professionals.
• Further research is essential to build trust and facilitate the adoption of interpretable AI in healthcare.