Estimating Cardiac Stroke Volume from the Seismocardiogram Signal

Executive Summary

This study investigates the feasibility of using seismocardiography (SCG) to estimate beat-by-beat cardiac stroke volume (SV) by extracting SCG features corresponding to cardiac events and comparing them to Doppler-derived SV measurements. A feed-forward neural network trained with backpropagation was employed, achieving an average correlation coefficient of 0.70 with Doppler estimates, with 95% of SCG-derived SV values falling within ±7.5 mL of Doppler values. The findings suggest SCG's potential as a non-invasive alternative for SV estimation, though further validation in diverse populations is needed.

“This study shows that heart vibrations measured on the chest (SCG) can estimate the amount of blood pumped by the heart (stroke volume) almost as accurately as ultrasound methods, using machine learning techniques.”

Answer Machine Insights

Q: What was the average correlation coefficient between SCG and Doppler stroke volume estimates?

The average correlation coefficient was 0.70.

In the best possible cases we had an average correlation coefficient of 0.70.

Q: What was the range of differences between SCG and Doppler stroke volume estimates?

The differences were within ±7.5 mL in 95% of cases.

Our preliminary results show that the estimated values, as in Table 1, in 95% of the cases, were in a 7.5 mL range of the values estimated by Doppler suprasternal measurement.

Key Results

The SCG-derived stroke volume estimates had an average correlation coefficient of 0.70 with Doppler-derived values.
95% of SCG-derived stroke volume estimates were within ±7.5 mL of Doppler-derived values.

Visual Evidence

Figure 1: Simultaneous SCG, Doppler and ECG signals. MC: Mitral valve close, AO: Aortic valve open, RE: Rapid systolic ejection, AC: Aortic valve closure MO: Mitral valve open [5 and 8].