Creating a Predictive Model of the Contrast Enhancement for Coronary CT Angiography by Using Statistical Analysis and Machine Learning

Abstract

The purpose of this study was to calculate statistically significant patient data and test bolus (TB) parameters in order to predict the contrast enhancement of main bolus (eMB) in coronary computed tomography (CT) angiography, and to create a predictive model of eMB with the calculated parameters by machine learning. A total of 126 patients underwent coronary CT angiography. Contrast material was administered at a fixed injection rate and volume. The peak enhancement (PE) and the time needed to reach peak (TP) of the TB were calculated for each patient. The dependency of MB contrast attenuation on these parameters was evaluated. Significant correlations were obtained among PE, TP, and the patient body surface area (BSA) with the eMB. The coefficient of determination of the linear regression model to estimate eMB by machine learning using the above three variables was 0.70 for the training data and 0.55 for the test data. For comparison, the coefficient of determination of the model using only BSA was 0.55 for the training data and 0.36 for the test data; the accuracy of the model created during this time was confirmed.