| Imaging diagnosis is an important means to detect liver tumors,which is important for early treatment of liver cancer patients.However,reading imaging diagnosis has brought too much workload to the radiologists,and the accuracy of diagnosis is limited by the subjective opinions of the doctor.Therefore,it is an inevitable trend of medical imaging development to automate detection of liver cancer in medical images.In recent years,convolution neural network(CNN)has been successful in a variety of computer visual tasks and has been used to the analysis and processing of medical images.However,the relevant researches mostly focus on CT images,while a little researches on MRI.CNN is used to the liver tumor detection task in this paper and the main works are as follows:(1)CNN based FasterRCNN object detection method was used to detect liver tumor, training data was augmented and the anchors used for region proposal were redesigned according to the liver tumor task,making the algorithm to outperformance in the liver tumor detection task.(2)This paper first proposed a context-relevant method based on the FasterRCNN method,which extracted features both from the surrounding of the liver tumor and the liver tumor and merged them into a context-based representation.This context-based representation fed to the liver tumor was more comprehensive and finally significantly improved the accuracy of the method.(3)A computer-aided diagnosis system was developed in this paper.The system do 3D modeling for the tissue and liver tumor in the MRI and show more specific and visual presentation to the doctor,assisting the doctor to locate the liver tumor quickly.Experiments on test data demonstrate that the method in this paper can effectively detect liver tumors in MRI images and context-based modeling for the surrounding of the liver tumor further improve the performance of the method. |
