Premium
Applied Predictive Modeling of Coronary Microvascular Disease using Coronary Doppler and Cardiac Echocardiography
Author(s) -
Patel Kishan U.,
Sunyecz Ian L.,
McCallinhart Patricia E.,
Bartlett Christopher W.,
Trask Aaron J.
Publication year - 2018
Publication title -
the faseb journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.709
H-Index - 277
eISSN - 1530-6860
pISSN - 0892-6638
DOI - 10.1096/fasebj.2018.32.1_supplement.843.17
Subject(s) - cardiology , coronary artery disease , medicine , coronary flow reserve , fractional flow reserve , blood flow , doppler echocardiography , coronary arteries , artery , diastole , myocardial infarction , coronary angiography , blood pressure
Type 2 diabetic (T2DM) coronary resistance microvessels (CRMs) show early inward hypertrophic remodeling and reduced wall stiffness, which are associated with decreased blood flow (CBF), coronary flow reserve, and disrupted flow patterns. These data may suggest that coronary microvascular disease (CMD) underlies the early pathophysiology of T2DM. Furthermore, distinct correlations have been identified between the Doppler coronary flow pattern in both normal and T2DM mice, potentially allowing for a predictive relationship between coronary flow patterns and CMD. Currently, CMD is extremely difficult to diagnose due to lack of non‐invasive methods. The goal of this study was to develop an applied predictive model for CMD non‐invasively in early T2DM using CBF combined with functional and structural measurements of the heart. Methods CBF of the left main coronary artery, aortic flow, E & A wave, left ventricular structural dimensions, and aortic diameter were recorded using high frequency, high resolution non‐invasive Doppler echocardiography (Vevo2100, Visual Sonics, Toronto, Canada) in 37 normal heterozygous Db/db mice and 36 T2DM homozygous db/db mice. Coronary flow patterns were analyzed using a Matlab program developed in our laboratory to identify 13 distinct parameters. All other flow patterns and structural and functional features were manually measured in the Vevo2100 software resulting in 13 additional parameters. All the parameters were feature engineered, and a factor analysis model was created and tested using various machine learning algorithms. Results Physiological data were subjected to a variety of machine learning models, and the “glmnet algorithm” in R software (generalized linear regression via penalized maximum likelihood) best predicted CMD with a cross validation accuracy of 80.19% and test dataset accuracy to date of 84.84%. Conclusion An applied predictive model for CMD in T2DM was developed using 73 mice, which shows a promising ability to accurately predict CMD in T2DM. Further studies will be undertaken to improve the predictive accuracy of the model to be suitable for clinical diagnostics and/or decision support. Support or Funding Information Supported by the National Institutes of Health (R00 HL‐116769 to AJT) and Nationwide Children's Hospital (to AJT and CWB). This abstract is from the Experimental Biology 2018 Meeting. There is no full text article associated with this abstract published in The FASEB Journal .