Deep Learning Identifies Digital Biomarkers for Self-Reported Parkinson's Disease

Area of Science:

• Neurology
• Digital Health
• Biomarker Discovery

Background:

• Traditional Parkinson's disease (PD) screening relies on conventional healthcare methods.
• Mobile health (mHealth) offers a potential independent method for PD detection and monitoring.
• Existing PD detection algorithms require improved generalizability using real-world data.

Purpose of the Study:

• To report the leading smartphone-based method for Parkinson's disease digital diagnosis from the DREAM challenge.
• To evaluate the efficacy of using accelerometer data for PD detection in real-world settings.

Main Methods:

• Utilized real-world accelerometer data from smartphone sensors.
• Applied 3D augmentation techniques to accelerometer records.
• Compared performance against state-of-the-art methods in the DREAM Parkinson's Disease Digital Biomarker Challenge.

Main Results:

• The developed smartphone-based method achieved an area under the receiver-operating characteristic curve of 0.87.
• Successfully differentiated Parkinson's disease patients from control subjects.
• Demonstrated significant improvement over existing state-of-the-art PD detection methods.

Conclusions:

• Smartphone-based methods show promise for accurate Parkinson's disease detection.
• This approach facilitates future at-home screening for PD and other movement-affecting neurodegenerative conditions.
• Real-world data collection enhances the generalizability of digital biomarkers for neurological disorders.