Fully automatic assessment of carotid artery curvature and diameter with non-invasive ultrasound

Abstract

An algorithm for performing fully automatic morphological analysis of the common carotid artery (CCA) in longitudinal ultrasound (US) B-mode scans is introduced. Arterial position, curvature and diameter are automatically estimated offline without human supervision. The CCA is automatically recognized by applying a method based on decimation, template matching and clustering to the first pair of US frames. This allows automatic placement of seed points in the CCA lumen. Next a sustain-attack filter (SAF) is applied in order to get raw estimates for CCA wall positions, used to deduct the CCA centerline in a region of interest (ROI). Arterial orientation and curvature are estimated by fitting a second order polynomial through the centerline. Subsequently the lines of sight in the ROI are vertically re-aligned based on the centerline so that the arterial walls shift to an approximately horizontal position, allowing subsequent lateral smoothing with horizontal kernels. This directional preprocessing attenuates speckle modulation without blurring perpendicularly to CCA walls. The SAF is then re-applied, yielding smooth delineation of both near and far wall. CCA diameter, orientation and curvature radius are calculated within the ROI. Automatic tracking of CCA movements is performed frame-by-frame. The method is tested on 30 US recordings (duration 6 seconds each) from 10 healthy volunteers. The algorithm successfully detects, without any user interference, CCA position, curvature and diameter for every subject considered. Comparison of results with and without curvature-compensated lateral smoothing shows significant decreases in the standard deviation of wall position estimates when the proposed pre-processing is applied. Correcting for local orientation at each line of sight yields the real CCA diameter, eliminating the error due to a non-perpendicular US beam orientation. © 2009 Springer Berlin Heidelberg.