Automated Plaque Detection and Agatston Score Estimation on Non-Contrast CT Scans: A Multicenter Study

TL;DR

This study addresses the burden of manual coronary plaque assessment on non-contrast chest CT by validating a 3D multiclass nnU-Net segmentation pipeline, integrated with TotalSegmentator organ masks, across three centers. Ground-truth plaques were manually labeled on 801 volumes and used to train and evaluate the model, achieving high detection performance and strong agreement with manual Agatston scoring after a linear correction ($slope=0.841$, intercept=$+16$ HU, $R^2=0.97$). The results show robust plaque detection (precision $0.893$, recall $0.891$, Dice $0.75$) and improved correlation relative to prior non-gated approaches, enabling accurate Agatston estimation on both gated and non-gated CT scans. This approach can enable scalable, opportunistic CAC screening from routine chest CTs, enhancing cardiovascular risk stratification in clinical practice.

Abstract

Coronary artery calcification (CAC) is a strong and independent predictor of cardiovascular disease (CVD). However, manual assessment of CAC often requires radiological expertise, time, and invasive imaging techniques. The purpose of this multicenter study is to validate an automated cardiac plaque detection model using a 3D multiclass nnU-Net for gated and non-gated non-contrast chest CT volumes. CT scans were performed at three tertiary care hospitals and collected as three datasets, respectively. Heart, aorta, and lung segmentations were determined using TotalSegmentator, while plaques in the coronary arteries and heart valves were manually labeled for 801 volumes. In this work we demonstrate how the nnU-Net semantic segmentation pipeline may be adapted to detect plaques in the coronary arteries and valves. With a linear correction, nnU-Net deep learning methods may also accurately estimate Agatston scores on chest non-contrast CT scans. Compared to manual Agatson scoring, automated Agatston scoring indicated a slope of the linear regression of 0.841 with an intercept of +16 HU (R2 = 0.97). These results are an improvement over previous work assessing automated Agatston score computation in non-gated CT scans.

Figures (5)

• Figure 1: Overall framework of the cardiac plaque detection algorithm. Given non-contrast CT scans, manual labelling segments plaques, while TotalSegmentator segments heart, aorta, and lung classes. Then, nnUNet generates 3D predictions from testing volumes.
• Figure 2: STARD Chart showing patient flow. In this graph, $n$ represents the number of scans.
• Figure 3: Agatston scores for the testing datasets. For the testing set, comparison of Agatston scores for automated and manual assessment showing (a) linear regression and (b) Bland-Altman plots.
• Figure 4: Examples of plaques (yellow arrows) in the coronary arteries and mitral valve in axial CT images.
• Figure 5: Examples of ground truth (top row) and corresponding predicted (bottom row) plaque burden measurements in axial CT images. For each example, the images are shown with labels and detections (red).