The Potential of Cardiac Function to Enhance the Machine Learning of Coronary Microvascular Disease
By Cole McLartyOn October 19, 2021, our Research Division continued our client-lead webinar series that, much like our blog series, strives to bring our community a venue where new scientific approaches, technologies and ideas can be shared.
| 11 Nov 2021
For this webinar, we heard from Dr. Aaron Trask, a Principal Investigator in the Center for Cardiovascular and Pulmonary Research at Nationwide Children's Hospital and an Assistant Professor in the Departments of Pediatrics & Physiology and Cell Biology at The Ohio State University College of Medicine. He trained as a physiologist and pharmacologist at Wake Forest University focusing on hypertensive heart disease and completed his postdoctoral training in cardiovascular physiology at Nationwide Children's Hospital focusing on coronary microvascular disease.
Dr. Trask's research program bridges these two dynamic and integrated tissues by studying processes involved in adverse micro and macro-vascular remodeling in the context of the cardiac cycle, including in early hypertension, type 2 diabetes, and metabolic syndrome. Current projects include elucidating the mechanisms of coronary and cardiac stiffness on coronary blood flow, heterocellular communication in coronary microvascular disease, and machine learning of coronary flow patterns as a direct method of predicting coronary microvascular disease. The latter was the topic of his webinar, “The Potential of Cardiac Function to Enhance the Machine Learning of Coronary Microvascular Disease” – watch in full here.
Below is a summary of the event:
According to the American Heart Association, Coronary Microvascular Disease (CMD) is an under-diagnosed subclinical causative factor to heart disease. It may precede the onset of coronary conduit disease and it is the dysfunction and adverse remodeling/biomechanics of the coronary microcirculation that impairs coronary blood flow in the presence or absence of coronary macrovascular disease. Risk factors contributing to CMD include hypertension, diabetes, obesity, metabolic syndrome, smoking and family history. Of particular interest to Dr. Trask’s lab is how this works with type 2 diabetes and metabolic syndrome. The challenge with these patients is that very early vascular remodeling, and the subsequent macrovascular changes, occurs mostly asymptomatically. This means that these people are unaware that there is a problem and that they will become patients in the future. Unfortunately, current diagnostic approaches are not overly specific. It is known that in animal models, impaired coronary flow is a hallmark of this disease progression, so Dr. Trask set out to see if flow model tracing, paired with machine learning, could be a powerful tool in this area.
Machine learning is a term used to explain the use of computer algorithms that can optimize themselves, automatically, when applied to large volumes of data. Through their “ability to learn”, these algorithms can adjust to meet new data conditions and, in this manner, improve their predictive power over time. As a branch of Artificial Intelligence, machine learning is used by large tech companies to various ends: Improving user engagement, reducing wait times, improving GPS guidance, and even fixing traffic jams. Machine learning is starting to be employed in the life science and medical spaces more broadly as well. Scientists and physicians take data obtained through scientific collection methods and push these data through computer algorithms, with the goal to improve early diagnosis of various conditions. Machine learning is however not a one size fits all approach to data processing and therefore to address his data, Dr. Trask broke down this work into the following steps:
- Coronary flow patterning.
- Automating data extraction from these patterns using MATLAB.
- Software development to “teach” it to recognize normal and diabetic flow patterns in the coronaries. Showing 70-80% accuracy at this stage.
- Improve the accuracy with factor analysis and machine learning, breaking the variables down in 6 different fields. Leading to further improvement in the predictive value of the program.
From all this work, it became clear that the model seemed to be working, but that they would need a much large ‘N’ value to improve their predictive power to verify the approach. Their next step was to develop a process that included data collection with our ADV500 pressure-volume loop equipment in conjunction with echocardiography in almost 500 subjects. The goal with their machine learning approach is to reach a predicative power of 90-95% in the clinic. As they are almost at this point, a second webinar will be scheduled in the upcoming period and Dr. Trasks will share the results with us. Stay tuned for that.
We would like to thank Dr. Aaron Trask for taking the time to join us for this talk. As we look towards our next webinar on November 16 - Whole-Organ Bioengineering: The Future of Transplantation Medicine – we hope that you can join us. Click here for more information related to our webinar series, to watch past events, register for future events or even to suggest topics for discussion.