AUTOMATIC SEGMENTATION OF LIVER TUMOUR USING A POSSIBILISTIC ALTERNATIVE FUZZY C-MEANS CLUSTERING

摘要

A knowledge-driven algorithm to automatically delineate the tumour region of human liver from computed tomography (CT) images for computer-aided liver diagnosis system is proposed in this paper. Automatic segmentation of liver tumours from computed tomography images is difficult due to the ambiguous nature of liver and tumour boundaries, the complicated appearance of tumours, the variation in the contrast of liver tissues and vessels, the different sizes and shapes of tumour, and the presence of many small metastases. Hence, the proposed algorithm first segments the liver and then extracts the tumour region from this segmented result. It uses a region-growing algorithm supported by pre-processing and post-processing for segmenting liver and a possibilistic alternative fuzzy C-means clustering technique for segmenting the tumour region. The proposed algorithm is assessed by comparing the automatic segmented tumour results with manual segmented results established by human experts for 25 CT data sets. The evaluation shows good segmentation based on performance measures like accuracy, sensitivity, specificity and precision.