Fusion Kernel Fuzzy C-Means Clustering and Improved Distance Regularized Level Set Evolution Model of Thyroid Nodules Segmentation

摘要

The high intensity of noise, low contrast and complex background of ultrasonic images are factors that affect the accuracy in thyroid nodules segmentation. The commonly used Kernel fuzzy C-means clustering (KFCM) algorithm generates low segmentation accuracy and the distance regularized level set evolution (DRLSE) model, with low evolution efficiency, cannot satisfy the requirement of highly accurate segmentation on the weak edges of the ultrasound images. In order to improve the accuracy of above mentioned algorithms, the new thyroid nodule segmentation algorithm, which combines the KFCM method with the improved DRLSE model, is proposed in this paper. The nodule region is roughly located by the application of KFCM algorithm and the result is used as the initial level set evolution contour, which overcomes the high sensitivity of DRLSE model to the initial contour. The level set parameters are calculated automatically using this algorithm which means manual setting is no longer required. Moreover, the new boundary stopping function is constructed and Gaussian regularization replaces the level set penalty term in the DRLSE model and will increase the evolution efficiency of its level set. Comparative experiments show that the new method proposed can efficiently increase the accuracy of segmentation of thyroid nodule areas regardless of high noise and blurred or weak edges.