Detecting Parkinsonian Tremor From IMU Data Collected in-the-Wild Using Deep Multiple-Instance Learning.

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Parkinson's Disease (PD) is a progressive neurological disorder affecting millions globally, with early detection crucial for effective management.
• Current methods for PD symptom detection often rely on controlled environments, limiting real-world applicability.
• Automated detection of PD symptoms like tremor, rigidity, and bradykinesia using accessible sensors is a significant research goal.

Purpose of the Study:

• To develop and validate a novel method for automatically identifying tremorous episodes associated with Parkinson's Disease.
• To enable the detection of PD symptoms using inertial measurement unit (IMU) signals captured by smartphone devices.
• To address the limitations of current approaches by enabling detection under free-living, in-the-wild conditions.

Main Methods:

• A Multiple-Instance Learning (MIL) approach was employed, treating subjects as bags of accelerometer signal segments.
• Deep feature learning was combined with a learnable pooling stage for end-to-end training.
• The method was validated on a new dataset of 45 subjects with in-the-wild accelerometer recordings.

Main Results:

• The proposed method demonstrated good classification performance in identifying tremorous episodes.
• The algorithm successfully navigated the noisy data characteristic of in-the-wild recordings.
• The findings suggest the method's efficacy for real-world PD tremor detection.

Conclusions:

• The developed method offers a promising solution for the automated, in-the-wild detection of Parkinson's Disease tremors using smartphone sensors.
• This approach has the potential to significantly improve early diagnosis and patient outcomes through timely interventions.
• Further research can build upon this foundation to expand automated PD symptom monitoring in daily life.