Automatic Segmentation of Dynamic Thyroid Ultrasound Images Based on An Improved DRLSE Model

Abstract

Aiming at the characteristics of dynamic thyroid ultrasound images with high noise, low contrast, uneven gray distribution, and little change in continuous frame images, this paper improves the Distance Regularized Level Set Evolution (DRLSE) model and applies it to dynamics Segmentation of thyroid ultrasound images, It is of great significance for the correct diagnosis of thyroid disease. The algorithm uses bilateral filtering instead of Gaussian filtering to smooth the image, which protects the edge information of the image while smoothing the noise. Improve the edge indicator function of DRLSE and use the step-by-step method for image segmentation, which improves the speed and accuracy of segmentation. In order to achieve automatic segmentation of dynamic ultrasound images, the centroid of the thyroid segmentation result of each image is used as the initial contour point of the next adjacent frame of image for segmentation. Experiments show that the algorithm solves the problems of poor anti-noise ability of the DRLSE model, long evolution time, and manual modification of the initial contour. It not only improves the speed of segmentation of the thyroid area but also improves the segmentation Precision.