Background: Carotid plaque volume (CPV) can be measured by 3D ultrasound and may be a better predictor of stroke than stenosis, but analysis time limits clinical utility. This study tested the accuracy, reproducibility, and time saved of using an artificial intelligence (AI) derived semiautomatic software to measure CPV ("auto-CPV").
Methods: Three-dimensional (3D) ultrasound images for 121 individuals were analyzed by 2 blinded operators to measure auto-CPV. Corresponding endarterectomy specimen volumes were calculated by the validated saline suspension technique. Inter-rater and intrarater agreement plus accuracy compared with the volume of the endarterectomized plaque were calculated. Measurement times were compared with previous manual CPV measurement.
Results: The mean difference between auto-CPV and surgical volume was small at (±s.d.) [95% confidence interval [CI]] 0.06 (0.24) [-0.41 to 0.54] cm3. The intraclass correlation (ICC) was strong at 0.91; 95% CI 0.86-0.94. Interobserver and intraobserver error was low with mean difference (±s.d.) [95%CI] 0.01 (0.26) [-0.5 to 0.5] cm3 and 0.03 (0.19) [-0.35 to 0.40] cm3 respectively. Both showed excellent ICC with narrow confidence intervals, ICC = 0.90; 95% CI (0.85-0.94) and ICC = 0.95; 95% CI (0.92-0.96). Auto-CPV measurement took 43% the time of manual planimetry; median (IQR) 05:39 (01:58) minutes compared to 13:05 (04:15) minutes, Wilcoxon rank-sum test, P < 0.01.
Conclusions: Auto-CPV assessment is accurate, reproducible, and significantly faster than manual planimetry. Improved feasibility means that the utility of CPV can be assessed in large population studies to stratify risk in asymptomatic carotid disease or assess response to medical treatment.
Copyright © 2023 The Authors. Published by Elsevier Inc. All rights reserved.