Research On Medical Organizations Textrue Recognition And Diagnosis Based On Neural Network

In this thesis, liver ultrasound images are recognised correctely based on applying computer technology, bio-medicine technology, digital image processing, and pattern recognition. The paper introduces pattern recognition methods of template matching, LMES and BP neural network to recognise liver ultrasound images.The analysis methods include the statistical method of gray level co-occurrence matrix and the signal processing method of Gabor wavelet which are used to abstract texture feature.Firstly, after analying the principle of liver ultrasonic and the gray information of liver B-scan ultrasonic images we known that the pathology of the liver B-scan ultrasound image rich in texture informations. There are a lot of methods about applications of pattern recognition in texture images: in this paper we use the gray level co-occurrence matrix and the Gabor wavelet to abstract the texture, because the gray level co-occurrence matrix can effectively abstract the texture's energy, entropy, local smooth and so on. And Gabor wavelet can be a multi-scale, multi -direction of the effective depiction of texture characteristics, and we try to find useful informations for liver ultrasound image classification, which will be applied to the identification of the normal liver, fatty liver and cirrhosis liver ultrasound images in three categories; secondly, we use lots of samples of the regional choosed image, and all three types with equal quantity samples. On the basis of texture characteristics which are extracted by gray level co-occurrence matrix and Gabor wavelet transform, we use the pattern recognition method to recognise the three types of liver B-scan ultrasound images, and chose three categories methods: templates category, geometric category and Neural network category, pratically we apply specific template matching, LMES and BP neural network. Finally, we compare the experiment to select the best texture feature extraction methods and the best pattern recognition classification, because of the different forms of data that are different from the scope, so the standardization of data is essential, normalized the value of tectrue characteristics extracted, then test;and then we make a full analysis and the specific details about the various types of texture characteristics extraction methods's effectiveness and the various categroy's performance, and evaluated the efficiency from the input to the process of identification. The experiments show that application of Gabor wavelet transform to extracte texture characteristics are more effective expression of texture's identity, and in the identification, the improved BP neural network to distinguish is with a higher rate, and it's rate reaches 90.3 percent to identify the normal liver, cirrhosis and fatty liver ultrasound images, it is meet the medical reguirment.In this paper, the study have greatly application significance, it can be used in clinic, it can improve the doctor's work efficiency and accurateness of diagnosis as an aid means of doctor's clinical diagnosis.