Toward a Speech-Based Model of Premanifest Huntington's Disease Progression Using Deep Neural Networks.

Area of Science:

• Neuroscience
• Computational Biology
• Speech Science

Background:

• Huntington's disease (HD) is a progressive neurodegenerative disorder with motor, cognitive, and psychiatric symptoms.
• Current staging methods are insensitive to early, premanifest changes.
• Speech abnormalities are potential digital biomarkers, but distinguishing pre-HD from controls is challenging.

Purpose of the Study:

• To assess the feasibility of a speech-only approach for discriminating pre-HD from healthy controls.
• To evaluate the effectiveness of various machine learning classifiers and feature sets.

Main Methods:

• Collected speech samples from 94 HD individuals (38 pre-HD) and 36 controls using a standardized protocol.
• Extracted 188 lexical and prosodic features.
• Trained four classifiers (random forest, SVM, XGBoost, DNN) using 10-fold cross-validation with different feature configurations.

Main Results:

• Deep neural networks (DNNs) using only the Caterpillar passage achieved 81% accuracy for pre-HD vs. controls.
• Accuracy increased to 83% for prodromal HD and 87% for all HD vs. controls.
• Additional features or tasks did not enhance classification performance.

Conclusions:

• A brief, structured speech task combined with deep learning accurately classifies pre-HD.
• Speech analysis is a scalable and objective tool for early HD detection and monitoring.