Research On B-Ultrasonic Liver Image Classifier

B-ultrasonic imaging is becoming more and more important in the liver diagnosis, with the enhancement and wild application of B ultrasonic instrument. However, as many subjective factors existing during the diagnosis process, such as visual fatigue, carelessness, and diagnosis level etc, not all doctors can detect liver accurately cancer, especially inchoate liver cancer. As these kinds of factors are hard to abstain, developing computer-aided diagnosis systems appears to be crucial not only for academic research, but also for clinical usage. For doctors, this CAD system can improve the accuracy of the diagnosis of liver cancer, estimate the possibility of suffering liver cancer, remind doctors or patients to get farther check and diagnosis, decrease the false diagnosis and false negative rate.According to the characteristic of B-ultrasonic liver image, this dissertation compares the classification effect of different classifiers with the fuzzy enhancement used in processing of images. The classifiers adopted K-means, back-propagation neural network support vector machine. The feature selection is also studied in this dissertation, because different features used in different classification methods may attain different result. The main achievements of this dissertation are as follows:(1)Fuzzy enhancement is applied in pre-processing of B-ultrasonic images, and the effect is studied with the three classifiers. The experiment results indicate that the usage of fuzzy enhancement method can attain better recognition rate of liver cancer for all the three classifiers.(2)25 features are extracted in this dissertation, and the recognition rate of different features combination are analysed in next step. Through experiments we find the optimal features combination.(3)This dissertation does the further research in compare between K-means, BP and SVM with different kinds of classification structures, and find out that the SVM possesses best recognition rate and stability. The parameters used in SVM are analysed with a series of experiments. How to set the corresponding parameter is concluded.