Neural Network-based Classification of Static Lung Volume States using Vibrational Cardiography^.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Non-invasive health monitoring enhances medical treatment delivery and efficiency.
• Vibrational cardiography (VCG) offers a non-invasive method for physiological signal acquisition.
• Understanding the interplay between cardiac cycles and respiratory states is crucial for advanced monitoring.

Purpose of the Study:

• To develop a neural network model capable of classifying lung volume states (high lung volume [HLV] and low lung volume [LLV]) from VCG signals.
• To investigate the feasibility of using cardiac cycle analysis for respiratory state detection.
• To establish a foundation for VCG-based cardio-respiratory monitoring systems.

Main Methods:

• Acquired 15,619 cardiac cycles from 50 subjects in both HLV and LLV states.
• Utilized a 1D convolutional neural network (CNN) architecture for classification.
• Trained and evaluated the CNN model using an 80/20 train/test split.

Main Results:

• The CNN model achieved a high classification accuracy of 99.4% on the testing dataset.
• Demonstrated the ability to differentiate between HLV and LLV states based on VCG data.
• Indicated a strong correlation between VCG signal characteristics and lung volume.

Conclusions:

• A CNN model can effectively classify lung volume states from VCG cardiac cycles.
• VCG analysis provides a viable non-invasive approach for respiratory monitoring.
• These findings support further research into VCG for integrated cardio-respiratory assessment.