Tumor Recognition in Liver CT Images Based on Improved CURE Clustering Algorithm

Spectral Computed Tomography (CT) images can help doctors diagnose the lesions of the organs and the types of organ lesions. According to the gray level information and spatial information of the spectral CT image of the liver, the characteristics of the image are selected. Using the improved Clustering Using Representatives (CURE) unsupervised clustering algorithm to cluster the image features to automatically identify liver tumors, not only does it not need to manually mark a large number of training samples, but also does not require long training on the classification model. This paper has two improvements to the CURE algorithm: (1) Liver is divided into multiple categories, and then combining the multiple categories into two categories according to certain rules instead of being divided into two categories directly by CURE. (2) When the liver in the spectral CT image is healthy, in order to meet the practical application, analyze the image before classification to avoid separating the normal liver into two categories. The experimental results show that the location of liver tumors is well marked based on the improved CURE clustering algorithm. It has a good clinical guidance value after being evaluated by clinicians and imaging doctors.