Machine Learning for Evaluating Vulnerable Plaque on Coronary Computed Tomography Using Spectral Imaging.

Abstract

Background: This study aimed to determine whether spectral imaging with dual-energy computed tomography (CT) can improve diagnostic performance for coronary plaque characterization.

Methods and results: We conducted a retrospective analysis of 30 patients with coronary plaques, using coronary CT angiography (dual-layer CT) and intravascular ultrasound (IVUS) studies. Based on IVUS findings, patients were diagnosed with either vulnerable or stable plaques. We computed 7 histogram parameters for plaque CT numbers in 120 kVp images and virtual monochromatic images ranging from 40 to 140 keV at 5-keV intervals. A predictive model was developed using histogram data of optimal energy, plaque volume or stenosis, and a combination of both. The model's performance was evaluated by calculating the area under the receiver operating characteristic curve (AUC) using 5-fold cross-validation. Peak diagnostic performances for each histogram parameter were observed at various energy levels (40-110 keV) in the univariate logistic regression model. The histogram model demonstrated optimal diagnostic performance at 65 keV, with an AUC of 0.81. The combined model, incorporating histogram data and plaque volume, achieved an AUC of 0.85, which was similar to the performance of qualitative CT characteristics (AUC=0.88; P=0.70).

Conclusions: Spectral imaging with dual-energy CT can enhance the diagnostic performance of machine learning using CT histograms for coronary plaque characterization.