The Tumor Texture Analysis And Its Clinical Application Based On Multi-scale Wavelet Transform

Texture is used to describe image gray-scale or color change, it is the repetition of specifictexture primitive, there is a lot of useful non-visual information implied in texture. Textureanalysis is to extract the characteristic parameters which represent the texture characteristicsinformation to quantitative or qualitative the image features. Using texture analysis techniqueprocessing the medical images to gain some characteristic is the high stage of medical imageprocessing. Texture analysis technology has got more and more application in medical imageprocessing, such as image recognition, image segmentation, image detection etc, texture hasbeen used in medical image aided in the quantitative analysis of diagnosis and so on.Malignant tumor has become a threat to human life, lung cancer accounts for30%of thewhole body malignant tumor, it has been one of the most common cancer in China. The18-FFDG PET (fluoro deoxy glucose positron emission tomography) can present themetabolism, growth, biological characteristics, drug taking non-invasive and quantitatively.18F-FDG PET can show the non-uniform FDG uptake, thus it can effectively quantify thedegree of tumor heterogeneity, the heterogeneity parameters is used to guide radiation andchemotherapy or predict the curative effect of radiation and chemotherapy, texture analysis hasbeen widely used in guiding tumor staging, diagnosis, judging recurrence.There are many kinds of methods to extract texture features, but each has its advantagesand disadvantages, there is no one method suit all the images, it is necessary to find a suitableextraction method to the specific image. Due to the multi-scale features of texture itself,Multi-scale wavelet transform won a good space in texture analysis for its goodmulti-resolution characteristics. The traditional pyramid wavelet transform decompose imageonly on the low frequency part, the information in the high frequency sub-band can’t be used.The gray level co-occurrence matrix (GLCM) is a classical statistical approach for the analysisof image texture features, but it is the overall analysis of the image processing, lack of local texture details information, it is difficult to study the mutual relationship between pixels.Because of the complexity of the tumor texture in PET images, it is necessary to researchthe rich texture details on a different scale and sub-band. The tumor has three-dimensionalspace characteristics itself, therefore, combining wavelet transform and gray levelco-occurrence matrix to complement their disadvantages can combine the local randomness andoverall regularity, thus we propose the three dimensional texture analysis based on multi-scaletechnique to extract PET texture, then the correlation studies is researched between the texturecharacteristic value and some clinical parameters.