Research And System Implementation On Extraction And Analysis Of Texture Features Of Thyroid MR Images

Medical image processing occupies an important position in medical diagnosis.Feature extraction is an important technique in medical image processing.The qualitative analysis of medical images can improve the recognition rate of medical images and assist doctors in making clinical image disease diagnosis.As an important characteristic of medical images,texture feature extraction technology has always been a hot topic in the field of medical image recognition.MR images have the advantages of clear imaging,high resolution and high contrast,so texture feature extraction is widely used in the analysis of such images.For texture feature extraction technology,which is sensitive to noise and insufficient information extraction,this thesis proposes a new gray-level co-occurrence matrix(GLCM)and local binary model for thyroid MR images.Pattern,LBP)texture feature extraction method,and developed a thyroid MR image texture feature extraction and recognition system to assist physicians in the timely and accurate diagnosis and treatment of diseases.The specific research work is as follows:(1)We studied the multi-scale GLCM window adaptive thyroid MR image texture feature extraction method.Firstly,the Gaussian pyramid is used to decompose the image into multi-scale space,which compensates for the defects of the single scale and effectively enhances the coarse-grained description of the image.Gradient detection is used to extract the contour information of the region of interest on a single scale.By setting the sliding window size and constructing GLCM based on the contour information,the GLCM window adaptive adjustment is implemented.This not only preserves the detail but also effectively suppresses the noise.Finally,the feature combinations of multiple scales are selected as the final feature vector through the feature selection.Logistic regression model was used to perform the two-classification experiment,which verified the effectiveness of the proposed method in extracting the texture information of thyroid MR images.(2)We studied the thyroid MR image texture feature extraction method based on high order directional Derivative of Mean Completed Local Binary pattern(DM_CLBP).On the basis of traditional LBP,second-order difference method is used to improve the calculation method of neighborhood pixel value difference,and the concave and convex information of the image is extracted effectively.The mean gray value of the pixel block is used instead of the gray value of a single pixel point to increase the correlation between the pixels,thereby improving the resistance of the texture feature to noise;The symbol and the amplitude of the complete local two value mode under the uniform mode were encoded and reorganized,which enhances the completeness of image information description.Logistic regression model was used to carry out the two-classification experiment,which verified the adequacy of the improved algorithm to characterize image texture information.(3)A computer-aided diagnosis system for thyroid MR image texture feature extraction and recognition was designed and developed.The system consists of two functional modules: image texture feature extraction and recognition,which can improve the accuracy of early diagnosis of thyroid cancer and reduce the rate of missed diagnosis and misdiagnosis.