Comparative Study of FCM, ARKFCM, K-Means and PSO Based Algorithms for Segmentation of Thyroid Nodules

Abstract

A thyroid nodule is an abnormal cell growth in the thyroid gland and seems as a palpable or non-palpable mass. Thyroid nodules are very frequent in clinical practice. Most thyroid nodules are benign, and only a small portion (5%-10%) of nodules are malignant. At present, high resolution ultrasonography is the principal imaging modality for diagnosing and classifying benign and malignant nodules. In order to visualize anatomic structures (tissues, body organs, nodules) of interest from medical images, segmentation plays an indispensable role. Precise segmentation of organs would allow accurate measurements, simplify visualization and, consequently, make the diagnosis more reliable. The aim of this study is to compare the performance of Fuzzy C-Means (FCM), K-means, Particle Swam Optimization (PSO) and Adaptive Regularized Kernel Fuzzy C-Means (ARKFCM) based segmentation techniques for accurate delineation of nodules using clinical thyroid ultrasound images. Preprocessing is carried out to remove speckle noise and to achieve better segmentation results. The results of these segmentation algorithms are analyzed based on six performance metrics Peak Signal to Noise Ratio (PSNR), Mean Square Error (MSE), Structural Similarity Index (SSIM), Normalized Cross Correlation (NCC), segmentation accuracy and time taken for segmentation. Experimental evaluation revealed that, PSO and ARKFCM segmentation algorithms outperformed compared to k-means and FCM algorithms.