Objective: The primary objective of this study was to develop and evaluate machine learning methods for the automatic prediction of standardized clinical scales using multimodal data from patients with Parkinson’s disease (PD) and Huntington’s disease (HD).

Background: Neurodegenerative diseases like PD and HD present significant global healthcare challenges, requiring early and accurate monitoring of motor, cognitive, and psychiatric symptoms. Integrating multimodal information; encompassing motor, oculographic, and vocal signals offers a path toward more objective, automated, and precise assessment of disease severity.

Method: Multimodal data were collected from 48 participants (45 with PD, 3 with HD) using mixed-reality (MR) goggles equippedwith inertial, oculographic, and vocal sensors. Participantscompleted 17 interactive tasks, including speech exercises, eye-tracking assessments, and motor activities. For each clinicalscale, optimal feature subsets were selected with lasso, Sequential Floating Forward Selection (SFFS), and SequentialBackward Floating Selection (SFBS). Predictive models (SVR and Random Forest) were trained for both single-scale and multi-output settings, employing late-fusion approaches such as stacking and Dempster–Shafer theory.

Results: The most effective approach for predicting clinical scalesinvolved multi-output models combined with late fusion. Fusion strategies improved prediction quality for 40 out of 43 analyzedscales, with R2 values reaching between 0.28 and 0.71. Specifically, the “Rising from a chair” scale achieved an R2 of 0.95 with a Mean Absolute Error percentage (MAE%) of 1.69%. Establishing a dedicated, fixed set of parameters for each scaleallowed for stable and satisfactory predictive performance.

Conclusion: Automated clinical score prediction using multimodal MR sensor data and late fusion provides a reliable, objectivealternative to traditional assessments. These ML-driven modelsdemonstrate the potential to monitor neurodegenerative diseaseprogression with accuracy comparable to clinical specialists, particularly for complex motor and vocal tasks.

« Back to 2026 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/multimodal-information-exploration-for-automated-clinical-scale-prediction-in-neurodegenerative-diseases/