High-sensitivity acceleration sensor detecting micro-mechanomyogram and deep learning approach for parkinson's...

Area of Science:

• Biomedical Engineering
• Neurology
• Sensor Technology

Background:

• Parkinson's disease (PD) diagnosis relies on clinical symptoms and imaging.
• Conventional sensors cannot detect subtle high-frequency muscle vibrations (micro-mechanomyogram or micro-MMG).
• High-sensitivity acceleration sensors offer enhanced vibration detection capabilities.

Purpose of the Study:

• To investigate high-frequency micro-MMG in the extensor pollicis brevis muscle of Parkinson's disease patients (PwPD) and healthy controls (HC).
• To assess the diagnostic potential of micro-MMG for differentiating PwPD from HC.
• To develop a deep learning model for PD classification using MMG data.

Main Methods:

• Development of high-sensitivity acceleration sensors capable of detecting vibrations >10 dB lower than commercial sensors.
• Measurement of low-frequency (MMG, <15 Hz) and high-frequency (micro-MMG, ≥15 Hz) vibrations in the extensor pollicis brevis muscle.
• Application of a deep learning model to classify PwPD and HC based on MMG and micro-MMG features.

Main Results:

• Detection of previously undetectable micro-MMG in PwPD and HC.
• Significant differences in micro-MMG frequency characteristics between PwPD and HC.
• Lower micro-MMG energy in PwPD compared to HC during muscle power output.
• Higher low-frequency MMG energy in PwPD compared to HC.
• A deep learning model achieved 92.19% accuracy in classifying PwPD and HC using both MMG types.

Conclusions:

• High-sensitivity sensors and micro-MMG analysis provide crucial information for distinguishing PwPD from HC.
• Micro-MMG represents a promising biomarker for Parkinson's disease.
• The developed deep learning model demonstrates the potential for a non-invasive diagnostic system for PD.