Speckle in ultrasound images: Friend or FOE?

Abstract

Contrary to the popular belief of treating speckle related pixels as noise and filtering an ultrasound image for speckle noise removal, the practical importance and use of these pixels in performing a multi-organ segmentation of the thyroid gland is studied in this research work. In this work, speckle related pixels are classified into three echogenic levels and then used to segment an ultrasound image of the thyroid gland into the trachea, carotid, muscles and thyroid. Novel techniques are introduced to estimate the anterior boundaries of the thyroid gland using low pass filtered intensity gradients of the hyperechoic speckle pixels in transverse and longitudinal ultrasound scans, respectively. An energy functional similar to active contour models is defined to segment that carotid artery using hypoechoic speckle pixels. The proposed technique was executed on 88 images of the thyroid gland. Clinical significance of using speckles to segment is determined by validating on 32 images of the thyroid gland by measuring the overlap with the Ground Truth segmentation obtained from two expert doctors using Dice coefficient as the overlap measure.