Cardiovascular Intelligence: Novel quantitative cardiology, deep learning and machine learning models to predict clinical outcomes
MetadataShow full item record
Arterial compliance has been recognized as a critical parameter in governing pulsatile flow dynamics. It has traditionally been assumed constant throughout the cardiac cycle. Its computation has been based either on the classic Windkessel model in diastole or the stroke volume over pulse pressure method in systole. We proposed a novel compliance-pressure loop (CPP loop) approach to quantify arterial compliance and compare it to existing linear and nonlinear methods. This was followed by an analysis of increased pulse wave reflections in hypertension due to impedance mismatching and how the effective energy transmission to the vasculature is compromised. Their quantification in the time and the frequency domains were compared. This was followed by working on quantifying the impedance mismatches. Augmentation index (AIx) is used to quantify the augmented systolic aortic pressure that impedes ventricular ejection. Its use as an index of wave reflections is questionable. We showed that AIx is quantitatively different from the reflection coefficient under varied physiological conditions. To effectively implement the novel CPP loop and calculate wave reflections with existing clinical equipment, we developed machine learning models to predict arterial diameters, reflection coefficient, and pulse wave velocity from radial pressure data. This was followed by more stand-alone deep learning and machine learning modeling of intracranial hypertension prediction in traumatic brain injury patients and survival outcome predictions in patients undergoing catheterization.