Interpretable machine learning for dementia: A systematic review.

Area of Science:

• Artificial Intelligence in Medicine
• Computational Neuroscience

Background:

• Automated dementia diagnosis using machine learning (ML) is growing but lacks clinical utility.
• Key challenges include developing robust, generalizable, and explainable ML models.
• Interpretable models and post hoc explainability methods are crucial for clinical trust.

Approach:

• A systematic literature review identified 92 studies on interpretable ML for dementia diagnosis.
• Analysis focused on classification performance, validation, and reporting standards.
• Evaluation of current state-of-the-art in explainable AI for neurodegenerative diseases.

Key Points:

• Studies show promising classification performance but exhibit variability in validation and reporting.
• Over-reliance on popular datasets limits generalizability.
• Understanding interpretability methods is critical for analyzing model explanations.

Conclusions:

• Future work requires clinician collaboration for validating explanation methods and inferring disease pathology.
• Patient-specific explanations are essential to demonstrate clinical benefits.
• Enhanced validation and reporting standards are needed for reliable AI in dementia care.