基于改进DRLSE模型的甲状腺3D超声图像自动分割

Automatic Segmentation of Thyroid 3D Ultrasound Images Based on An Improved DRLSE Model

从超声图像中准确分割甲状腺区域是甲状腺疾病手术计划的关键之一。本文一方面,针对甲状腺超声3D图像,提出利用边缘指示函数和面积项系数改进的距离正则化水平集演化(Distance Regularized Level Set Evolution,DRLSE)模型来实现甲状腺区域的有效分割;另一方面,根据3D超声图像相邻帧之间甲状腺变化较小的特点,通过计算已分割图像的质心,作为相邻帧图像分割初始点来实现3D图像的自动分割。实验表明,采用本文改进DRLSE模型分割甲状腺3D超声图像,平均分割精度可以达到90%以上。

Accurate segmentation of the thyroid region from ultrasound images is one of the keys to a surgical plan for thyroid disease. On the one hand, this paper proposes utilizing an improved edge indicator function and the area regular coefficient to improve the Distance Regularized Level Set Evolution (DRLSE) model to achieve effective segmentation of the thyroid region. On the other hand, according to the characteristics of small thyroid change between adjacent frames of the three-dimensional ultrasound image, the centroid of the segmented image is used as the initial point for the adjacent frame image. Thus the segmentation of the three-dimensional image is performed automatically. Experiments show that the improved DRLSE model can segment the thyroid three-dimensional ultrasound image effectively, and the average segmentation accuracy is about 90%.