Research And System Implementation Of Feature Extraction And Recognition Of Liver CT Images

Feature extraction,a key technology in medical image processing which plays an important role in computer aided medical diagnosis,has attracted much attention of a large number of scholars in recent years and become a research hotspot in medical image processing field.Efficient feature extraction algorithm can provide powerful support for early detection and treatment of diseases through extracting key information from medical images to represent the regions of interest to achieve the auxiliary diagnosis.Furthermore,in clinical examination,CT imaging which is widely used because of the high spatial resolution and high signal-to-noise ratio is one of the medical imaging techniques commonly used in the diagnosis of hepatocellular carcinoma.Therefore,it is of great significance to conduct researching on feature extraction methods for liver CT images.At present,the feature extraction algorithms of medical images are mainly based on texture and shape,which is lack of representation of high-level image semantic information.In addition,there are some deficiencies that the description of information is inadequate in low-level features.According to these problems,the feature extraction algorithm is researched based on liver CT images in this thesis.The main contents are as follows:(1)An improved multi-scale local binary mode(MSLBP)feature extraction algorithm for liver CT image is proposed to alleviate the problems of traditional local binary mode method that multi-scale features cannot be extracted efficiently and a large number of high-order neighborhood pixel information in liver CT images may be lost.This algorithm achieves the purpose of describing the texture features of liver images from different scales by making full use of the information of high-order sampling points and averaging the neighboring pixels,which can highlight the relationship between neighboring pixels.The effectiveness of the improved MSLBP algorithm is verified by multiple comparative experiments.(2)A multi-feature fusion feature extraction algorithm for liver CT images based on deep learning is proposed in order to improve the defect that only the details but not high-level semantic information can be expressed by bottom features.The algorithm finally obtains a feature vector that can comprehensively describe the image information by fusing the high-level semantic features obtained by deep learning with the bottom features achieved by traditional feature extraction methods.The experimental results show that the classification and recognition accuracy of liver CT images can be effectively improved.(3)A system for liver CT image feature extraction and recognition is designed and developed,which mainly consists of two functional modules,image feature extraction and image recognition.This system can realize the purpose of normal and abnormal liver images classification by classifying the images based on the features obtained through different extraction and analysis operations,which can assist doctors in diagnosis to a certain extent.