Automated Quantification of 3-D Carotid Ultrasound Vessel Wall Texture Change Predicts Vascular Events.

Abstract

Objective: To reduce reliance on labor-intensive and operator-dependent explicit plaque segmentation, we evaluated whether a texture-change score (ΔTexture) derived from semi-automated 3-D ultrasound (3-DUS) vessel wall segmentation predicts vascular events better than baseline vessel wall texture (BL-Texture), vessel wall volume change (ΔVWV) and Framingham risk score.

Methods: 3-DUS images were acquired from 272 patients with carotid atherosclerosis at baseline and 1 y later. Vessel walls were delineated using a validated, minimally user-initialized segmentation workflow. Longitudinal texture change was summarized as vessel wall ΔTexture and compared with BL-Texture, ΔVWV and Framingham risk score. Patients were followed for up to 5 y (median, 3.16 y) for myocardial infarction, transient ischemic attack and stroke.

Results: ΔTexture and BL-Texture predicted vascular events (log-rank, p < 0.001 for both), whereas ΔVWV did not. No event occurred in the lowest-risk tertile of ΔTexture. The area under the receiver operating characteristic curve of ΔTexture was 0.89, compared with 0.82 for BL-Texture, 0.53 for ΔVWV and 0.53 for Framingham risk score. In Cox regression, ΔTexture (median hazard ratio, 4.05; p < 0.001) and BL-Texture (median hazard ratio, 3.63; p < 0.001) were significant predictors, whereas ΔVWV and Framingham risk score were not.

Conclusion: Vessel wall ΔTexture derived from minimally user-initialized 3-DUS vessel wall segmentation predicted vascular events better than ΔVWV and Framingham risk score. This approach provides a scalable imaging biomarker for vascular event risk assessment without requiring explicit plaque segmentation.