Detecting Parkinsons' symptoms in uncontrolled home environments: a multiple instance learning approach.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Neurology

Background:

• Parkinson's Disease (PD) motor symptom detection is crucial for patient management.
• Current methods often rely on supervised learning, which struggles with the imprecise data from home monitoring.
• There's a need for non-invasive, in-home monitoring systems to track symptom fluctuations and medication effects.

Purpose of the Study:

• To develop a weakly supervised machine learning framework for automatic detection of Parkinson's Disease motor symptoms in daily living environments.
• To enable continuous, out-of-clinic monitoring for tracking medication cycles and providing clinical feedback.
• To overcome the limitations of coarse ground truth data in home-based monitoring settings.

Main Methods:

• Formulated symptom detection with incomplete ground truth as a multiple instance learning (MIL) problem.
• Employed a MIL algorithm based on axis-parallel rectangle (APR) fitting in the feature space.
• Utilized data from five triaxial accelerometers worn by two Parkinson's Disease patients over four days each.

Main Results:

• The trained MIL detector successfully identified symptom-prone time windows and localized symptom instances.
• Detected subject-specific symptoms, such as dyskinesia, in Parkinson's Disease patients.
• Results aligned with daily symptom logs maintained by the patients, validating the system's accuracy.

Conclusions:

• Weakly supervised learning, specifically MIL, is effective for detecting Parkinson's Disease motor symptoms using in-home monitoring data.
• The developed system provides a feasible approach for tracking medication cycles and offering clinical insights outside laboratory settings.
• This technology holds promise for improving the management and understanding of Parkinson's Disease progression in real-world environments.