To systematically review explainable AI (XAI) techniques for speech-based detection of Alzheimer’s disease and mild cognitive impairment, emphasizing the systematic nature of the review.
Key Findings:
XAI methods like SHAP, LIME, and attention mechanisms were utilized in the studies, achieving AUC values ranging from 0.76 to 0.94, indicating good model performance. Consistent identification of acoustic markers (pause patterns, speech rate) and linguistic features (vocabulary diversity, pronoun usage) was observed.
Interpretation:
XAI techniques show potential for enhancing clinical interpretability of AI models in cognitive decline detection, but significant gaps exist in stakeholder engagement and real-world validation, which are crucial for effective implementation.
Limitations:
Limited stakeholder engagement in the development of XAI methods, need for real-world validation of XAI techniques, lack of standardized evaluation frameworks for XAI in cognitive assessment, and the necessity for diverse populations in validation studies.
Conclusion:
While XAI techniques enhance the interpretability of AI models for detecting cognitive decline, further research is needed to address existing gaps and improve clinical adoption, highlighting the importance of stakeholder involvement.
