Ultrasound Image Data Mining Based Research On Key Technology In HIFU Non-invasive Monitor

With its advantages of non-invasive, no radiation and low cost, ultrasound guided HIFU is supposed to be a promising therapy method for tumor. As one of the key problems in HIFU related technique, ultrasound-based temperature and HIFU effect estimation have been attracting many researchers attention. But there is still no resolution which is mature enough for clinic use.By now, most ultrasound signal or image-based non-invasive temperature or HIFU effect estimation methods are based on ideal sound field and need to know some sonic characters of tissue first. That is a very difficult condition to achieve. Data mining technology whose aim is analyzing data from different perspectives and summarizing it into useful information, has its advantages in medical data analysis and research This thesis concentrates on: Dada mining parameters from ultrasound images which can help to monitor HIFU therapy effect; Finding the rules and giving the understandable and acceptable evaluation of HIFU effect. Temperature is a very important parameter need to be monitored during HIFU therapy. Data mining temperature related ultrasound image texture parameter is based on the hypothesis that texture will change with tissue temperature. Gray Level Co-Occurrence Matrix (GLCM) , box fractal dimension, wavelet methods were used to extract texture parameters from ultrasound images at different temperature and their subtraction with ultrasound image at basic temperature(37℃). Results show that original ultrasound images' texture has no obvious relationship with temperature of tissue, but some GLCM texture parameters and wavelet coefficient energy parameters of subtraction images have linear relationship with temperature of tissue from 40℃to 80℃. Eight parameters have linear relationship with temperature were used to construct principal components analysis based multiple regression prediction model. The regression formula's precision is 3℃at temperature under 70℃.Another way to estimate HIFU effect is monitoring lesion of tissue. A sub-pixel ultrasound image based method is proposed in this thesis and correlation distance was defined for quantization of featured points' movement and change in HIFU lesion area image. Results show that temperature has no quantitative relationship with lesion degree, neither dose, but sub-pixel cross-correlation vector field can reflect the ablation lesions position and correlation distance is helpful for detecting the degree of the beam ablation lesions. Correlation distance based SVM classification method can help to detect HIFU lesion degree automatically and efficiently.Besides research on monitoring temperature and HIFU lesion, another theme of this thesis is to find a way which can automatically provide the HIFU lesion area's shape and position after therapy. Data mining edge information of ROI (region of interest) from high-noise ultrasound images is the key. We proposed a double zero level set, double speed approach to extract the organ and lesion area's contour. Gray level change was used to initial zero set automatically Serial "slice" contours were used to reconstruct three dimension (3D) HIFU lesion areas in the sample. Result: Experiments show that level set contour extraction method based 3D reconstruction is helpful in monitoring the size, shape and location of HIFU lesion.