From Black Box to Biomarker: Sparse Autoencoders for Interpreting Speech Models of Parkinson's Disease

TL;DR

This paper uses sparse autoencoders to interpret speech models for Parkinson's disease, revealing interpretable features linked to clinical biomarkers and improving understanding of disease-related speech deficits.

Contribution

It introduces a novel mask-based activation for sparse autoencoders, enabling interpretable representations from small biomedical speech datasets.

Findings

Identified speech features associated with PD, such as spectral flux and flatness.

Linked spectral flux to MRI-based putamen measurements.

Demonstrated potential for biomarkers in disease monitoring.

Abstract

Speech holds promise as a cost-effective and non-invasive biomarker for neurological conditions such as Parkinson's disease (PD). While deep learning systems trained on raw audio can find subtle signals not available from hand-crafted features, their black-box nature hinders clinical adoption. To address this, we apply sparse autoencoders (SAEs) to uncover interpretable internal representations from a speech-based PD detection system. We introduce a novel mask-based activation for adapting SAEs to small biomedical datasets, creating sparse disentangled dictionary representations. These dictionary entries are found to have strong associations with characteristic articulatory deficits in PD speech, such as reduced spectral flux and increased spectral flatness in the low-energy regions highlighted by the model attention. We further show that the spectral flux is related to volumetric measurements of the putamen from MRI scans, demonstrating the potential of SAEs to reveal clinically relevant biomarkers for disease monitoring and diagnosis.