Design And Implementation Of Multi-Phase Hepatic CT Image Based CAD System

CAD (Computer Aided Diagnosis) is a system which uses medical imaging and image processing technology combined with artificial intelligence and pattern recognition to assist radiologist to detect lesions in time and positioning precisely and improve the diagnosis accuracy ultimately. As it were, CAD is the radiologist's"third eye"while the use of CAD System can significantly improve the diagnosis accuracy.Liver cancer is a very treacherous one in malignant tumor. In the world, the incidence of liver cancer and the mortality rate is very high. Most of the liver cancer patients always have no symptoms or signs. Even if some of the patients have any signs, it is hard to attract attention of patients, so the patients are in a bad way when they find. We always use surgery, interventional therapy, radiotherapy and treatment by Chinese herbs to save the patients. Because the treatments will bring great pain to patient, and difficult to cure, early detection of liver cancer is particularly important. In fact, the shape of liver lesions on imaging tends to diffuse, there is no fixed shape. Different lesions on unenhanced CT images are often quite similar, so it is hard to distinguish the early lesions of the liver by naked eye. Clearly, CAD has a major significance for early detection of liver cancer.In clinical medicine, the contrast agent has been proved to be unharmful to humans, and has been widely used. Multi-phase CT scans can show the timing characteristics of liver lesions, so it can provide more information about the disease to be more effective in helping radiologist in diagnosis than non-enhanced CT scans. Because abdominal CT images includes not only the liver organ, as well as other abdominal organs, this article uses the watershed algorithm to segment the liver from the abdominal CT to study. We use wavelet transform to extract regions of interest in automatic detection. In feature extraction, scattergram is a concentrated expression of two images and sensitive about small changes between two images, so it has many inherent advantages in image feature extraction. In the designing of classifier module, we convert a 4 classes classifying problem into 3 binary classify problems by using artificial neural network. Finally, we obtained the best classification accuracy of 0.9797, 0.9851 and 0.9753 for normal-abnormal, cyst-otherdisease and carcinoma-haemangioma sub problems respectively. Diagnosis accuracy and stability have been greatly improved.The main works and innovation points of this thesis are described as below:1. Improved watershed algorithm is used to segment the liver from the abdominal CT to study. We have achieved good results.2. SIFT algorithm is used to registry the CT image, so it can make the multi-phase CT image match well.3. Gabor wavelet transform is used to automatically detect the regions of interest. Wavelet transform can effectively detect the ROI.4. Scattergram is used to extract the features of ROI. Scattergram can use the features of the multi-phase CT images effectively and it can improve the diagnosis accuracy. Diagnosis accuracy and stability have been greatly improved.