Recognizing Dementia from Neuropsychological Tests with State Space Models

TL;DR

This paper introduces Demenba, a state space model-based framework for automatic dementia detection from neuropsychological test speech, demonstrating improved accuracy and scalability over prior methods with potential for enhanced transparency.

Contribution

Demenba is a novel, scalable state space model for automatic dementia classification that outperforms previous approaches and benefits from fusion with large language models.

Findings

Outperforms prior approaches in dementia classification by 21%

Scales linearly in memory and computation with sequence length

Gains additional improvement when fused with large language models

Abstract

Early detection of dementia is critical for timely medical intervention and improved patient outcomes. Neuropsychological tests are widely used for cognitive assessment but have traditionally relied on manual scoring. Automatic dementia classification (ADC) systems aim to infer cognitive decline directly from speech recordings of such tests. We propose Demenba, a novel ADC framework based on state space models, which scale linearly in memory and computation with sequence length. Trained on over 1,000 hours of cognitive assessments administered to Framingham Heart Study participants, some of whom were diagnosed with dementia through adjudicated review, our method outperforms prior approaches in fine-grained dementia classification by 21\%, while using fewer parameters. We further analyze its scaling behavior and demonstrate that our model gains additional improvement when fused with large language models, paving the way for more transparent and scalable dementia assessment tools. Code: https://anonymous.4open.science/r/Demenba-0861