Determining the Respiratory State From a Seismocardiographic Signal--A Machine Learning Approach
Executive Summary
This study investigates the relationship between respiration and seismocardiographic (SCG) signals using machine learning models, including multiple regression analysis (MRA), support vector regression (SVR), and neural networks (NN). SCG features such as fiducial point amplitudes, timings, and frequency components were extracted from signals collected from 18 healthy subjects. The NN model demonstrated the highest predictive accuracy with Pearson correlation coefficients of 0.908 and 0.904 for mitral closure (MC) and aortic opening (AO) time points, respectively, suggesting SCG's potential for respiratory state segmentation and clinical applications in cardiopulmonary monitoring.
Answer Machine Insights
Q: What was the most accurate model for predicting respiratory amplitudes?
The neural network model was the most accurate, achieving Pearson correlation coefficients of 0.908 for MC and 0.904 for AO.
The NN model scored the highest Pearsons correlation coefficient (r = 0.908), which means there was an almost linear relationship between the predicted respiration and the actual respiration.
Q: What features were used to predict respiratory states from SCG signals?
Features included fiducial point amplitudes, timings, and frequency components such as power, mean frequency, and median frequency.
Timings and amplitudes were included as features since these changes during respiration. The power, mean- and median frequencies of the signal between MC and MO are also used as features.
Key Results
Neural network achieved Pearson correlation coefficients of 0.908 (MC) and 0.904 (AO).
Sum of squared errors for NN predictions were 11.71 (MC) and 12.05 (AO), outperforming SVR and MRA.
Clinical Snapshot
Evidence Rating
Relevance
high Priority