Carotid Artery Wall Segmentation in Ultrasound Image Sequences Using a Deep Convolutional Neural Network

Intima-media thickness (IMT) of the common carotid artery is routinely measured in ultrasound images and its increase a marker pathology. Manual measurement being subject to substantial inter- intra-observer variability, automated methods have been proposed find contours intima-media complex (IMC) deduce IMT thereof. Most them assume that these are smooth curves passing through points with strong intensity gradients expected between lumen intima, media adventitia layers. These assumptions may not hold depending on image quality arterial wall morphology. We therefore relaxed developed region-based segmentation method learns appearance IMC from data annotated by human experts. This deep-learning uses dilated U-net architecture proceeds as follows. First, shape location identified full-image-height patches using original resolution. Then, actual performed at finer spatial resolution, distributed around thus identified. Eventually, predictions combined majority voting segmented region extracted. On public database 2676 accuracy robustness outperformed state-of-the-art algorithms. The first step was successful $$98.7\%$$ images, overall mean absolute error estimated $$100\pm 89\,\mu $$ m.