论文部分内容阅读
The liver cancer,a hazard to people's health,is common in china.How to prevent the liver cancer has become an important task of tumor study,of which the key issue is the early-stage diagnosis of the liver cancer.With its unique advantages as no radiation,real time,repeatability,low cost and expedience to operate,the ultrasound diagnosis has become an effective measure for diagnosing the space-occupying liver lesions.The ultrasound clinical diagnosis,however,relies on doctors' unaided viewing and estimating,thus the diagnosis results heavily depend on diagnosis physicians' clinical experience,resulting in probably misdiagnosis to the nature of the space-occupying lesions.An objective,timely and accurate computer-aided diagnosis is urgently necessary.Along with the development of the computer science,computer-aided diagnosis has been applied in many medical fields in recent years.However,the research on the recognition of the liver space-occupying lesions in ultrasound images is still at its early stage.In this paper,the recognition of liver space-occupying lesions in ultrasound images is studied and developed,associated with the clinic diagnosis,a computer-aided diagnosis effective and easy to operate is worked out.Experiments are done on a total of 240 cases of liver images,including 60 cases of normal liver images,60 cases of liver cancer images,60 cases of liver hemangioma images and 60 cases of liver cyst images.Firstly,combining the request of clinic diagnose for computer-aided diagnosis,the ROI are extracted for each image from the standpoint of the criterion by which clinicians diagnose liver space-occupying lesions.Experiments mainly extract the gray and the texture feature,and a rectangle is selected in both the ROI and the normal area,respectively.Twenty-one numeric features are extracted in each image according to the selected ROI.Secondly,the structure of back-propagation artificial neural network for training is determined as three-level BP neural network.The suboptimum feature vector of each level of the network is acquired.Then,the feature selection is conducted based on the structure of the network and the function of each level of network,while considering the serial number curve of the features.Finally,the samples of each network will be trained and the acquired weights and thresholds corresponding to each level of network are saved after training.The computer-aided diagnosis system is generated by applying the saved weights and thresholds each level of network,consequently realizing diagnosis and recognition of the space-occupying lesion on ultrasound images by less handwork.The test samples include 52 cases of normal liver images,40 cases of liver cancer images,53 cases of liver hemangioma images and 65 cases of liver cyst images.The test result is as follows: 52 cases of normal liver images are all exactly recognized,with the recognition rate being 100%; for 65 cases of liver cyst images,two are recognized as liver cancer,with the recognition rate being 96.9%; for 53 cases of liver hemangioma images,three are recognized as liver cancer,with the recognition rate being 94.3%; 40 cases of liver cancer images are all exactly recognized,with the recognition rate being 100%.