First in Human Experience of a Novel No-implant Interatrial Shunt Device for Heart Failure with Preserved Ejection Fraction
HFSA ePoster Library. Barker C. 09/10/21; 343395; 169
Colin Barker
REGULAR CONTENT
Login now to access Regular content available to all registered users.
Abstract
Discussion Forum (0)
Introduction: Seismocardiography (SCG) is a potential tool to assess the clinical status of heart failure (HF) patients. Preliminary data suggests that certain features extracted from SCG signal may be utilized to predict HF readmission, which in turn may help identify patients requiring more intensive therapy or monitoring. Machine learning may be utilized for identification of potential SCG features to optimize readmission predictive value.
Objective: The objective of this study was to utilize machine learning to identify SCG features that correlate with HF readmission prediction.
Methods: SCG recordings were longitudinally collected from 40 HFrEF patients, among them 9 were readmitted during the 12 m study period. One hundred and three features were extracted from SCG waveforms including waveform variability, amplitude, spectro-temporal and heart rate variability parameters. Some of the features involved separating SCG waveforms into clusters that correlated with breathing phases. Four machine learning methods (information gain, random forest, ANOVA and “ReliefF”) were used to select optimum features.
Results: Features identified by various machine learning methods as most predictive of HF readmission are shown in Table 1 with the performance measure for each feature shown in parenthesis. The listed features are: IFrms,avg=mean of the root mean square (RMS) of instantaneous frequency of the 2 “cluster medoids.” WVbC=variability of SCG waveform before clustering. WViC and WVIC=intra and inter cluster waveform variability. SCG2amp(0.5-50),avg,SCG2amp(20-50),avg=mean RMS amplitude (normalized to SCG peak-to-peak) around SCG2 in 0.5-50 & 20-50 Hz bands, respectively. PEP=pre ejection period. SCG1amp(0.5-50),avg=mean RMS amplitude around SCG1 in 0.5-50 Hz. MFdiff=mean frequency difference between clusters.
Conclusions: SCG instantaneous frequency, waveform variability, pre-ejection period, and SCG amplitudes were the most predictive of HF readmission in this initial analysis. More data are needed to confirm consistency of these results. If successful, the proposed approach may have utility for management of HF patients.
Objective: The objective of this study was to utilize machine learning to identify SCG features that correlate with HF readmission prediction.
Methods: SCG recordings were longitudinally collected from 40 HFrEF patients, among them 9 were readmitted during the 12 m study period. One hundred and three features were extracted from SCG waveforms including waveform variability, amplitude, spectro-temporal and heart rate variability parameters. Some of the features involved separating SCG waveforms into clusters that correlated with breathing phases. Four machine learning methods (information gain, random forest, ANOVA and “ReliefF”) were used to select optimum features.
Results: Features identified by various machine learning methods as most predictive of HF readmission are shown in Table 1 with the performance measure for each feature shown in parenthesis. The listed features are: IFrms,avg=mean of the root mean square (RMS) of instantaneous frequency of the 2 “cluster medoids.” WVbC=variability of SCG waveform before clustering. WViC and WVIC=intra and inter cluster waveform variability. SCG2amp(0.5-50),avg,SCG2amp(20-50),avg=mean RMS amplitude (normalized to SCG peak-to-peak) around SCG2 in 0.5-50 & 20-50 Hz bands, respectively. PEP=pre ejection period. SCG1amp(0.5-50),avg=mean RMS amplitude around SCG1 in 0.5-50 Hz. MFdiff=mean frequency difference between clusters.
Conclusions: SCG instantaneous frequency, waveform variability, pre-ejection period, and SCG amplitudes were the most predictive of HF readmission in this initial analysis. More data are needed to confirm consistency of these results. If successful, the proposed approach may have utility for management of HF patients.
Introduction: Seismocardiography (SCG) is a potential tool to assess the clinical status of heart failure (HF) patients. Preliminary data suggests that certain features extracted from SCG signal may be utilized to predict HF readmission, which in turn may help identify patients requiring more intensive therapy or monitoring. Machine learning may be utilized for identification of potential SCG features to optimize readmission predictive value.
Objective: The objective of this study was to utilize machine learning to identify SCG features that correlate with HF readmission prediction.
Methods: SCG recordings were longitudinally collected from 40 HFrEF patients, among them 9 were readmitted during the 12 m study period. One hundred and three features were extracted from SCG waveforms including waveform variability, amplitude, spectro-temporal and heart rate variability parameters. Some of the features involved separating SCG waveforms into clusters that correlated with breathing phases. Four machine learning methods (information gain, random forest, ANOVA and “ReliefF”) were used to select optimum features.
Results: Features identified by various machine learning methods as most predictive of HF readmission are shown in Table 1 with the performance measure for each feature shown in parenthesis. The listed features are: IFrms,avg=mean of the root mean square (RMS) of instantaneous frequency of the 2 “cluster medoids.” WVbC=variability of SCG waveform before clustering. WViC and WVIC=intra and inter cluster waveform variability. SCG2amp(0.5-50),avg,SCG2amp(20-50),avg=mean RMS amplitude (normalized to SCG peak-to-peak) around SCG2 in 0.5-50 & 20-50 Hz bands, respectively. PEP=pre ejection period. SCG1amp(0.5-50),avg=mean RMS amplitude around SCG1 in 0.5-50 Hz. MFdiff=mean frequency difference between clusters.
Conclusions: SCG instantaneous frequency, waveform variability, pre-ejection period, and SCG amplitudes were the most predictive of HF readmission in this initial analysis. More data are needed to confirm consistency of these results. If successful, the proposed approach may have utility for management of HF patients.
Objective: The objective of this study was to utilize machine learning to identify SCG features that correlate with HF readmission prediction.
Methods: SCG recordings were longitudinally collected from 40 HFrEF patients, among them 9 were readmitted during the 12 m study period. One hundred and three features were extracted from SCG waveforms including waveform variability, amplitude, spectro-temporal and heart rate variability parameters. Some of the features involved separating SCG waveforms into clusters that correlated with breathing phases. Four machine learning methods (information gain, random forest, ANOVA and “ReliefF”) were used to select optimum features.
Results: Features identified by various machine learning methods as most predictive of HF readmission are shown in Table 1 with the performance measure for each feature shown in parenthesis. The listed features are: IFrms,avg=mean of the root mean square (RMS) of instantaneous frequency of the 2 “cluster medoids.” WVbC=variability of SCG waveform before clustering. WViC and WVIC=intra and inter cluster waveform variability. SCG2amp(0.5-50),avg,SCG2amp(20-50),avg=mean RMS amplitude (normalized to SCG peak-to-peak) around SCG2 in 0.5-50 & 20-50 Hz bands, respectively. PEP=pre ejection period. SCG1amp(0.5-50),avg=mean RMS amplitude around SCG1 in 0.5-50 Hz. MFdiff=mean frequency difference between clusters.
Conclusions: SCG instantaneous frequency, waveform variability, pre-ejection period, and SCG amplitudes were the most predictive of HF readmission in this initial analysis. More data are needed to confirm consistency of these results. If successful, the proposed approach may have utility for management of HF patients.
Code of conduct/disclaimer available in General Terms & Conditions
{{ help_message }}
{{filter}}