An Improved K-Means Clustering for Segmentation of Pancreatic Tumor from CT Images

Pancreatic tumor is a deadly disease in which cancerous (malignant) cells form in the tissues of the pancreas. There are numerous types of pancreatic cancers, the most common is pancreatic adenocarcinoma. Almost 90% cases suffer from pancreatic adenocarcinoma. However detecting lesions from medical images is very challenging, given that the shape of the pancreas in abdomen is very irregular, low contrast in edges, variability in location which requires more complex and accurate segmentation of boundary of pancreas from abdomen images. In computer-aided diagnosis system, automatic segmentation of a specific organ and its tumor is very important. The proposed framework provides a way to apply K-means clustering method (unsupervised learning algorithm) on pancreas CT image to detect the area of interest from the background. Our aim is to build an efficient framework to segment tumors from pancreas CT images which is most commonly used in the domain of medical imaging as well as help clinicians in better decision making for surgical planning. It is observed from the results that K-means clustering segmentation algorithm produces more accuracy on CT imaging datasets. Early detection of pancreatic adenocarcinoma helps in immediate treatment planning for individuals.