Segmentation-free Approach to the Classification of the Strain of the Carotid Plaque from Ultrasound Images
Date
Authors
Victoire, Marina
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Carotid atherosclerosis is an arterial disease caused by the accumulation of plaque inside the walls of the carotid artery, a major artery that carries oxygen-rich blood from the heart to the brain. The buildup of plaque inside the artery wall leads to narrowing, or stenosis, which could reduce blood flow to the brain. Rupture of atherosclerotic plaque and resulting embolization places the patient at an increased risk of stroke. Determining plaque composition can help determine which patients are at higher risk of having a stroke. High risk plaques have been shown to have a thin fibrous cap and a lipid rich necrotic core [22]. One method to determine the composition of the plaque is to measure the amount of strain, and hence, deformation that the plaque endures during a cardiac cycle. Large deformation of plaque are indicative of softer lipid-rich necrotic tissues whereas minimal deformation tend to indicate harder, mostly calcified or fibrotic tissues. Under the strain of repeated cardiac cycles (heartbeats), plaques may rupture, and small pieces of fibrofatty emboli or blood clots (thrombi) may separate from the plaque and lodge themselves in smaller arteries in the brain leading to a stroke. Ultrasound imaging is a non-invasive, widely available medical tool that has become an attractive imaging modality for initial evaluation of carotid artery stenosis. In previous research [1], images from ultrasound videos of thirteen patients diagnosed with carotid stenosis were processed to measure the amount of deformation of the plaque from one ultrasound image to the next using an algorithm called ‘optical flow’. In each image, the carotid plaque was manually segmented before applying the optical flow. In this thesis, we demonstrate that only the first image of each video needs to be segmented and that fitting a bounding box around the area of interest and propagating that bounding box to the rest of the images produce similar results, which greatly reduces the need for tedious manual segmentation. After applying the optical flow algorithm to the area within the bounding box, a clustering algorithm is applied to classify the patients into two distinct groups: (1) a low-risk group where there is minimal deformation of the plaque tissues (less strain) and, (2) a high-risk group where there is large deformation of the plaque tissues (more strain). These classifications of the elasticity of the carotid plaque provide an invaluable initial step to assess which patients might benefit from further diagnosis to better assess the risk of embolization.
Description
Keywords
Plaque deformation, Optical flow, Carotid plaque strain, Ultrasound, Segmentation, Classification