Abstract 13674: Optimal Hounsfield Threshold for Lipid-Rich Plaque by Artificial Intelligence-Enabled Quantitative CT Using Near-Infrared Spectroscopy

Abstract

Introduction: By near-infrared spectroscopy (NIRS), plaques with 4-mm maximum lipid-core burden index (maxLCBI4 mm ) ≥ 400 are lipid-rich plaque. By CCTA, the optimal Hounsfield threshold (HU) for lipid-rich plaques defined by presence of low-density non-calcified plaque (LD-NCP) is still unclear. Hypothesis: Artificial intelligence-enabled quantitative CT (AI-QCT) will enable identification of HU for lipid-rich plaques when compared to NIRS reference standard. Methods: 128 atherosclerotic plaques from 47 patients were enrolled prospectively. MaxLCBI 4mm (cutoff was defined as 400) and LD-NCP derived from AI-QCT (Cleerly LABS, Denver CO) were compared. LD-NCP using 3 different thresholds (<30, <50, and <70 HU) were analyzed by sensitivity, specificity, positive and negative predictive value, accuracy, and area Under the Curve (AUC). Results: Amongst 128 atherosclerotic plaques evaluated, 26 (20.3%) met NIRS criteria for lipid rich plaque (maxLCBI 4mm ≥400). The volume of LD-NCP was measure using three different thresholds, <30, <50, and <70 HU as well as criteria of PR > 1.1 and LD-NCP volume >1.8 mm 3 (LD-NCP volume threshold identified from AUC). This yielded a sensitivity, specificity, positive and negative predictive value, accuracy, and AUC of 92%, 88%, 67%, 98%, 89%, and 0.944 at <30HU, 96%, 79%, 53%,99% 82%, 0.921 at <50HU, and 96% 72%, 46%, 99%, 77%, and 0.907 respectively at <70HU (30 vs. 50 HU: p=0.008; 30 vs. 70HU: p=0.002 (Figure 1). Conclusions: A <30 HU threshold for definition of LD-NCP on AI-QCT correlates optimally with NIRS proven lipid-rich plaques.