Deep learning predicts cardiac output from seismocardiographic signals in heart failure

Comprehensive Summary

The aim of this study was to develop a deep learning model that can accurately predict cardiac output using data generated from a seismocardiogram (SCG), an electrocardiogram (ECG), and body mass index. A population of 83 individuals was used to develop this accurate model. Each had a simultaneously recorded SCG, ECG, and RCH (right heart catheterization), the demographic for each person was also known. Trial axial signals were used to capture inner axis mechanical motion, and an ECG was taken to provide precise temporal alignment of the cardiac events.There were two test models created: one using only SCG and ECG results and another using both of those results along with BMI. The model accounting for BMI proved to be more accurate. This model developed had a result of only 0.07 L/min less than the reference method for determining cardiac output, with a 95% confidence interval of -0.35 to +0.48 L/min. This model proved to be successful and had a high accuracy in patients with low cardiac output. The results of this study also indicated the potential use of SCG for earlier identification and therefore management of patients with impaired perfusion and can be used for noninvasive cardiac monitoring.

Outcomes and Implications

The current methods of determining cardiac output are right heart catheterization, which is very time consuming with many procedural risks. Additionally ECGs and MRIs have been used, but these are not entirely accurate due to confounding factors. The development of the deep learning model provides a more efficient while still accurate way for determining cardiac output. This model can be extremely beneficial in hospitals where invasive monitoring is not easily accessible. However, the authors note that the model should be tested for a larger population to ensure its accuracy.