Computer-Aided Liver Tumor Segmentation Research Based On CT Images

Nowadays, Primary liver cancer is a disease with high mortality rate due to the small amount of nerves inside the liver organ. Therefore, a slight liver disease or damage is difficult to be found out by human observers, and the initial cancer in liver has a long incubation period with relative smaller volume, no obvious clinical symptom, and so on. These features make it very difficult for a doctor to undertake visual diagnosis of liver cancer in early stage even with the help of some medical imaging tools such as Computed Tomography (CT), especially to the young residents who usually make missed diagnosis because of less clinical experience. As a result, tumor diagnosed as cancer is always in the middle-late stage, which leads to great difficulty in treatment of the liver cancer and the cure rate is extremely low. Therefore, developing an effective computer-aided diagnosis system to help doctor to diagnosis liver disease is essential and significant in clinical practice.This paper describes a novel method for the malignant liver tumor segmentation in the computer-aided diagnosis system as the following three parts:(1) Registration of soft organs in abdominal CT phased images, including Phase-Only Correlation (POC) rigid registration and Thin-Plate Spline (TPS) non-rigid registration, showing that17landmarks selected in the liver contour with TPS method for non-rigid registration can achieve satisfactory registration result.(2) Aiming at overcoming the disadvantage of the manually selected landmarks, we proposed a novel method of selecting landmarks automatically based on liver binary image. The new detecting landmarks method can find out a number of landmarks to meet the registration requirements, and the experimental results show that the automatically detecting landmarks method is effective and can be used in the medical image registration.(3) For the experiments of liver tumor segmentation, the edge detection and subtraction methods are applied to the extraction of liver tumor candidate; a spherical score filter is proposed for detecting small liver cancer and metastasis; and further false-positive (FP) is eliminated by using texture feature-contrast, which is generated by the gray level co-occurrence matrix, or co-operating with other shape or intensity features.