Atherosclerotic Plaque Burden on Abdominal CT: Automated Assessment With Deep Learning on Noncontrast and Contrast-enhanced Scans

Abstract

Background: Abdominal aortic atherosclerotic plaque burden may have clinical significance but manual measurement is time-consuming and impractical.

Purpose: To perform external validation on an automated atherosclerotic plaque detector for noncontrast and postcontrast abdominal CT.

Materials and methods: The training data consisted of 114 noncontrast CT scans and 23 postcontrast CT urography scans. The testing data set consisted of 922 CT colonography (CTC) scans, and 1207 paired noncontrast and postcontrast CT scans from renal donors from a second institution. Reference standard data included manual plaque segmentations in the 137 training scans and manual plaque burden measurements in the 922 CTC scans. The total Agatston score and group (0-3) was determined using fully-automated deep learning software. Performance was assessed by measures of agreement, linear regression, and paired evaluations.

Results: On CTC scans, automated Agatston scoring correlated highly with manual assessment (R² = 0.94). On paired renal donor CT scans, automated Agatston scoring on postcontrast CT correlated highly with noncontrast CT (R² = 0.95). When plaque burden was expressed as a group score, there was excellent agreement for both the CTC (weighted kappa 0.80 ± 0.01 [95% confidence interval: 0.78-0.83]) and renal donor (0.83 ± 0.02 [0.79-0.86]) assessments.

Conclusion: Fully automated detection, segmentation, and scoring of abdominal aortic atherosclerotic plaques on both pre- and post-contrast CT was validated and may have application for population-based studies.

Keywords: 3D-UNet; Agatston score; Image processing; machine learning.

Figure 1.

STARD Chart showing patient flow.

Figure 2.

Agatston scores for CT colonography and renal donor datasets. (A, B) For CT colonography, comparison of Agatston scores for automated and manual assessments showing (A) linear regression and (B) Bland-Altman plots. (C, D) For renal donors, comparison of Agatston scores for noncontrast and post-contrast datasets showing (C) linear regression and (D) Bland-Altman plots. For (A, C), linear regression equation and R² are shown. For (B, D), bias and 95% limits-of-agreement are shown. For (C), full and partial range of scores are shown.

Figure 3.

Examples of concordant plaque burden measurements on midline sagittal reformatted CT images. (A) 74 year old man in CT colonography dataset and (B) 62 year old woman in renal donor dataset. In (B), (left pair) noncontrast and (right pair) post-contrast images are shown. For each example, the images are shown without and with the automated detections (green). In (A), Agatston group 3 was recorded for both manual and automated measurements. In (B), Agatston group 3 was recorded for both automated noncontrast and post-contrast measurement.

Figure 4.

Examples of discordant plaque burden measurements on midline sagittal reformatted CT images. (A) 66 year old man in CT colonography dataset and (B) 33 year old woman in renal donor dataset. For the renal donor example, (far left) noncontrast and (right three) post-contrast images are shown. For each example, the images are shown without and with the automated detections (green). In (A), some detections occurred below the automatically-determined aortic bifurcation level (yellow line). Consequently, while the manual Agatston group was 3, the automated group was 2. In (B), false positive detections occurred on the left ureter; there are neither plaque nor detections on the aorta. Consequently, while the automated noncontrast Agatston group was 0, the automated post-contrast group was 3.