Fair Cognitive Impairment Detection Through Unlearning

2026-06-16

AI Summary

This study addresses the fairness issue in mild cognitive impairment (MCI) detection models based on spontaneous speech, where performance disparities across subpopulations arise due to reliance on demographic cues. To mitigate this, the authors propose a novel multimodal framework that integrates speech, text, and image modalities and, for the first time, incorporates a gradient reversal–based forgetting mechanism to actively remove task-irrelevant demographic information from shared embeddings. Evaluated on the multilingual TAUKADIAL and PREPARE benchmarks, the proposed approach significantly outperforms state-of-the-art multimodal models, achieving high classification accuracy while substantially reducing performance gaps across gender and language subgroups. This advancement enhances both model fairness and cross-dataset generalization capability.

Abstract

Mild Cognitive Impairment (MCI) is a medical condition characterized by a noticeable decline in memory, language, or thinking abilities. MCI detection from spontaneous speech is promising for scalable screening. However, learned models often exploit demographic cues correlated with labels, resulting in a large performance gap across subgroups. We present a multimodal framework that combines (i) cross-model fusion between modalities (speech, text, and image), and (ii) unlearning using gradient reversal that discourages the shared embedding from encoding task-irrelevant demographic attributes. Evaluated on the multilingual benchmarks TAUKADIAL and PREPARE, our method outperforms the state-of-the-art multilingual and multimodal baseline in MCI classification while substantially reducing the performance gap across patient subgroups (sex and language). We further analyze transfer across datasets, showing that demographic unlearning helps learn more robust representations for MCI detection.

Problem

• Fairness
• Mild Cognitive Impairment
• Demographic Bias
• Spontaneous Speech
• Subgroup Performance Gap

Innovation

• unlearning
• multimodal fusion
• fairness
• gradient reversal
• MCI detection