Automated IMT estimation and BMI correlation using a low-quality carotid ultrasound image database from India

摘要

This paper presents AtheroEdgeLowRes (AELR), an extention of AtheroEdge™ from AtheroPoint™, and a solution to carotid ultrasound IMT measurement in low-resolution and overall low quality images. The images were collected using a low-end ultrasound machine during a screening study in India. We aim to demonstrate the accuracy and reproducibility of the AELR system by benchmarking it against an expert Reader's manual tracing and to show the correlation between the automatically measured intima media thickness (IMT) and the subjects' cardiovascular risk factors (i.e. body mass index - BMI). We introduced an innovative penalty function (PF) to our dual-snake segmentation technique, necessary due to the low image resolution. We processed 512 images from 256 patients, and correlated the AELR IMT values with the patients' age and BMI. AELR processed all 512 images, and the IMT measurement error was 0.011±0.099 mm with the PF correction and 0.173±0.127 mm without. AELR IMT values correlated with the Reader's values (r = 0.883) and also correlated with the subject's BMI and age. The AELR system showed accuracy and reproducibility levels that make it suitable to be used in large epidemiological and screening studies in emerging countries.