Measurement of carotid intima-media thickness (cIMT) by ultrasound (US) is recommended for reclassifying subjects at intermediate cardiovascular risk. As carotid US is a standard, cost-effective procedure, we explored whether the same B-mode images where cIMT is measured contain additional information of vessel wall microstructure.Methods
We developed a predictive model based on B-mode images acquired by two certified radiologists with a linear multi-frequency (7-10 MHz) transducer at the far wall of the common carotid artery in 41 young adults with normal cIMT. The subjects examined were 20 healthy controls with cIMT 0.36-0.63 mm and 21 hyperlipidemic adults with cIMT 0.38-0.59 mm (Fig.1a). Images were delineated and processed by Automated Quantitative Ultrasound Analysis (AQUA) to extract invariant image features. Discriminative analysis between healthy and hyperlipidemic was computed based on six invariant image features. Robustness was evaluated by leave-one-out cross validation (LOOCV) to estimate prediction power.Results
Discriminative analysis resulted in 100% classification between healthy and hyperlipidemic. LOOCV showed maximum accuracy of 71% and area under the curve (AUC) of 0.63 and classification based on thresholding of cIMT resulted in a maximum accuracy of 54% and AUC of 0.42, as expected from sample's selection criteria (Fig.1b).Conclusions
This study suggests that acoustic properties of lipids compared to elastin, collagen and water produce contrast in US images of hyperlipidemic subjects that can be detected by invariant image features. Further research is required to achieve a stable predictive model. Our preliminary results show that invariant image features contain additional, heretofore hidden information beyond cIMT.