A medical percussion instrument using a wavelet-based method for archivable output and automatic classification

Area of Science:

• Medical instrumentation
• Acoustic signal processing
• Machine learning in healthcare

Background:

• Medical percussion, a diagnostic technique since 1761, lacks a standardized instrument.
• Reproducible generation and accurate classification of percussion sounds are crucial for consistent diagnosis.

Purpose of the Study:

• To develop and validate a novel instrument for standardized medical percussion.
• To create a system capable of reproducible sound generation and accurate classification of percussion sounds into three distinct classes.

Main Methods:

• A push-pull solenoid plessor and polyvinyl chloride plessimeter generated percussion impulses.
• Acoustic signals were acquired using a high-speed data acquisition card and microphone.
• Signal processing and classification were performed using a complex Morlet wavelet and a MobileNetv2 convolutional neural network.

Main Results:

• The developed instrument reproducibly generated percussion sounds.
• The MobileNetv2 classifier achieved high accuracy in classifying sounds (e.g., 94.0% cross-subject test accuracy).
• Subject-specific and cross-subject accuracies demonstrated the system's robustness.

Conclusions:

• The developed instrument provides a standardized approach to medical percussion.
• The automated classification system shows significant promise for clinical applications.
• This technology can enhance the reliability and consistency of physical examinations.