Validation of Freezing-of-Gait Monitoring Using Smartphone

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Freezing of gait (FOG) is a significant motor symptom in Parkinson's disease (PD), characterized by walking blocks and reduced step length.
• FOG negatively impacts patients' quality of life and substantially elevates injury risk, necessitating careful monitoring.
• Previous FOG detection methods required multiple body-worn sensors and external computing devices, utilizing less sophisticated, hand-crafted features.

Purpose of the Study:

• To develop and validate a novel, smartphone-based system for detecting freezing of gait (FOG) in Parkinson's disease (PD) patients.
• To leverage a convolutional neural network (CNN) for analyzing gait data collected from smartphone sensors.
• To provide a more accessible and less intrusive method for continuous FOG monitoring in daily life.

Main Methods:

• Utilized smartphone-embedded accelerometer and gyroscope data from 32 PD patients during daily activities.
• Processed motion signals into the frequency domain, creating 2D images as input for a specialized CNN model.
• Validated the CNN model's performance for FOG detection.

Main Results:

• The proposed CNN-based system achieved high diagnostic accuracy.
• Achieved an average sensitivity of 93.8% and a specificity of 90.1% in discriminating FOG events from normal walking.
• Demonstrated superior performance compared to previously reported FOG detection settings.

Conclusions:

• The developed smartphone-based system enables precise and continuous monitoring of freezing of gait (FOG) without the need for cumbersome external sensors.
• This unconstrained sensing approach can significantly aid Parkinson's disease (PD) patients in effectively managing their condition in daily life.
• The findings highlight the potential of integrating advanced machine learning techniques with ubiquitous mobile technology for improved chronic disease management.