Objective: This study investigates the potential of mixed reality (MR) tool for multimodal data collection from individuals with Parkinson’s disease and examines the use of deep learning and machine learning techniques for disease classification and regression.

Background: Parkinson’s disease is a neurodegenerative condition that affects motor function, speech, eye movements, and posture. Mixed reality (MR) technology, equipped with sensors, cameras, and microphones, enables convenient multimodal data collection in home-based settings, reducing patient burden and stress.

Method: Missing values were first imputed using KNN and refined with Predictive Mean Matching. Data were denoised, normalized to controls, and features filtered by low variance and high correlation. Significant features were selected via LightGBM with SHAP values and stability selection using Elastic Net and LightGBM. Classification (PD vs. HC) employed Logistic Regression and LightGBM with nested GroupKFold CV. The MR-NeuroScore was computed from stable features, and regression models predicted 12 clinical scales, including Hoehn & Yahr, Berg Balance, MDS-UPDRS, and Timed Up & Go.

Results: Logistic Regression and  LightGBM and Elastic Net effectively classified Parkinson’s  achieving superior discrimination (AUC 0.99 vs. 0.98). The aggregated MR-NeuroScore achieved a hazard ratio close to 3, confirming its value as a holistic disease indicator. Performance across 12 clinical scales varied due to differences in effective target ranges, with the Berg Balance Scale (MAE below 9.5) and Timed Up & Go (MAE below 11.00) showing the strongest results.

Conclusion: The study showed that mixed-reality proved highly effective for capturing high-dimensional, heterogeneous data. Their analysis, combined with tailored strategies for handling missing values handled via a novel neural network, yielded robust results. Advanced data processing then allowed the selection of the most informative and stable features, forming the basis of the MR-Score for subsequent diagnostics.

References: The project was funded by The National Centre for Research and Development, Poland under Lider Grant no: LIDER/6/0049/L-12/20/NCBIR/2021 and by the Ministry of Science and Higher Education (MNiSW) under the project no. MNiSW/2025/DPI/53, Support for Students in Enhancing Their Competencies and Skills.

« Back to 2026 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/quantitative-neurodegenerative-risk-modelling-and-clinical-score-prediction-using-mixed-reality/