Ultrasonic Image Segmentation Algorithm of Thyroid Nodules Based on DPCNN

Abstract

The segmentation of ultrasound images of thyroid nodules is a key technology for computer-aided diagnosis of thyroid. How to achieve precise segmentation of nodules has always been a hot issue in the field of medical image segmentation. To solve the problem that the traditional models are sensitive to the background area when segmenting ultrasound images with low contrast, we propose an ultrasonic image segmentation algorithm for thyroid nodules based on pulse coupled neural network with direct current component (DPCNN) in this paper. Firstly, the algorithm performs rough location of suspicious region on the optimal segmentation image output by DPCNN iteration, and uses the comprehensive judgment criteria of the maximum variance and covariance of the local region to determine the lesion area. On this basis, the nodule image is segmented based on DPCNN according to the gray features of the nodule image, so as to realize the precise segmentation of the thyroid nodule area. The experimental results show that this algorithm can effectively achieve the accurate segmentation of thyroid nodule area and has good robustness.