Digital Biomarkers for the Assessment of Motor Symptoms in Parkinson’s Disease: From Daily Life to Intervention Evaluation

Abstract

The objective and continuous assessment of motor symptoms in Parkinson’s disease (PD) remains limited by the episodic nature of clinical evaluations. This work presents a dual monitoring methodology developed within the BioCliTe project, integrating both standardized MDS‑UPDRS Part III exercises and an ecologically valid daily‑life task. Data were collected using smartwatches that acquired accelerometer and gyroscope signals at 50 Hz in supervised and free‑living settings, guided by a mobile application that enabled automatic labelling. The recordings were processed through a reproducible pipeline including filtering, segmentation, windowing, and feature extraction in both time and frequency domains. Explainable machine‑learning models, such as decision‑tree ensembles, logistic regression, and SVM, were trained using interpretability methods (LIME, SHAP) to define digital biomarkers of tremor, bradykinesia, and gait. These biomarkers demonstrated strong capability to differentiate PD patients from healthy controls and to reflect motor severity in unsupervised environments. Results confirm the feasibility of diagnosing and monitoring PD symptoms outside clinical facilities through wearable‑based biomechanical analysis. Notably, the free-living task yielded a low‑cost and reproducible bradykinesia biomarker with robust performance in clinical and remote conditions. The defined digital biomarkers establish the basis for the EVINTERS project, aimed at evaluating therapeutic effects on symptom progression. This approach supports more personalized, continuous, and patient‑centered management beyond point‑in‑time assessments.

Keywords: Parkinson’s Disease, Digital Biomarkers, Wearable Devices

DOI: 10.54941/ahfe1007490