Recognition of Abnormal Chest Compression Depth Using One-Dimensional Convolutional Neural Networks.

Area of Science:

• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Evaluating chest compression depth (CCD) traditionally relies on sensors like accelerometers or gyroscopes.
• This method is prone to measurement errors from sensor inaccuracies and environmental interference.
• Existing techniques often require auxiliary computing devices, limiting practical application.

Purpose of the Study:

• To develop a more accurate and streamlined method for evaluating CCD effectiveness.
• To reduce the need for auxiliary computing devices in CCD assessment.
• To mitigate errors caused by sensor limitations and environmental factors.

Main Methods:

• Proposed a one-dimensional convolutional neural network (1D-CNN) classification model.
• Trained the 1D-CNN using pre-collected sensor signal data classified by chest compression criteria.
• Evaluated the model's performance against other CNN models and support vector machines using 937 labeled CCD results.

Main Results:

• The 1D-CNN model achieved an accuracy rate exceeding 95% in recognizing CCD results after sufficient training.
• The model effectively classifies sensor signal data, bypassing the need for direct displacement measurements.
• Execution time analysis indicated a balance between accuracy and hardware requirements, suitable for portable devices.

Conclusions:

• The proposed 1D-CNN model offers a highly accurate and efficient solution for evaluating chest compression depth.
• This approach minimizes the impact of sensor errors and environmental interference.
• The model's design is suitable for integration into portable medical devices, enhancing real-time feedback and usability.