Incorporation of Modified Region Growing into FCM Clustering for Extraction of Tumors in MR Images

Abstract

Due to the tremendous growth of medical Magnetic Resonance (MR) images, it becomes an essential requirement for an automated extraction of diagnostically salient parts of the images to refine the transmission and storage process. Research method: This paper proposes an automated unsupervised machine learning segmentation technique named fuzzy c-means clustering based on modified region growing algorithm (RG-FCM) for the extraction of tumors from MR brain images. The average intensity value of the foreground region extracted by the Otsu thresholding technique is selected as seed point of the region growing algorithm, and spatial constraints extracted from modified region growing technique are then incorporated into the objective function of Fuzzy C-Means clustering to improve cluster separation and refine centroid selection, particularly in images with uneven illumination. Results and Discussion: The empirical evaluation exhibits the superior performance of the presented technique, and performs better than Possibilistic Fuzzy C-Means (PFCM) and conventional FCM. The mathematical results evaluated on a dataset of 60 MR images indicate the improvement in Jaccard and dice indices by (3-5) and (6-8) % from PFCM and conventional FCM respectively. Therefore, this incorporation of the modified region growing technique into FCM can be employed for real time processing due to its less execution time and can be expanded for the computer-assisted identification of abnormal tissue proliferation in MRI brain images.