Towards Dys-XAI: Influence-Based Explanations for Dysarthria Severity Assessment

2026-06-19

AI Summary

This study addresses the poor interpretability of deep learning models in automated dysarthria assessment and the limited clinical intelligibility of existing explanation methods. To this end, the authors propose an influence-based, instance-level interpretability framework that, for the first time, introduces gradient-driven influence analysis into dysarthria evaluation. By efficiently approximating the influence score of each training sample on model predictions, the method identifies supportive and competitive reference cases, yielding explanations that are both auditable and clinically interpretable. Controlled deletion experiments demonstrate that removing just 5%–20% of high-influence samples induces systematic shifts in model predictions, thereby validating the reliability and practical utility of the proposed explanations.

Abstract

Dysarthria severity assessment is essential for therapy planning and longitudinal monitoring, yet manual perceptual rating is time-consuming and variable across clinicians. Although deep learning models achieve strong performance, their black-box nature limits clinical adoption. Existing speech explainability methods typically provide acoustic feature importance scores that are difficult for end-users to interpret. We propose an influence-based, instance-level explainability framework that explains each decision through supportive and competing training samples. Using gradient-based influence approximations, we compute per-utterance influence scores to identify supportive and competing training samples for each prediction. Controlled deletion experiments from 5 to 20 percent validate the explanations, showing that removing highly influential samples systematically shifts predictions. This approach provides auditable explanations by linking decisions to perceptible reference cases.

Problem

• Dysarthria severity assessment
• Explainable AI
• Influence-based explanation
• Speech explainability
• Clinical interpretability

Innovation

• influence-based explanation
• instance-level interpretability
• dysarthria severity assessment
• auditable AI
• gradient-based influence