Ultrasound Image Processing Based On Contourlet Transform For Phased HIFU System

High Intensity Focused Ultrasound (HIFU) has been explored for its therapeuticuse in the treatment of tumors. The main advantage of HIFU is its non-invasive nature,and the focus where therapy occurs can be placed deep within a patient’s bodywithout affecting the intervening tissue layers. The localized effects of HIFU and itsaccurate focusing capability make it an attractive non-invasive surgical modality. Inthe interactive image-guided HIFU therapy, fast and precise target localization is veryimportant for treatment planning. Ultrasound image guidance of HIFU therapy hasbeen used because of its portability, low cost, real time, simple integration with HIFUinstruments. Therefore, the use of US visualization for the guidance and monitoring ofHIFU therapies most often relies on the performance of the US image processing.Contourlet transform was introduced as a discrete domain multiresolution andmultidirection expansion using non-separable filter banks, and developed as a “true”two-dimensional representation for images. Contourlet is considered to be the newgeneration of wavelet in two and higher dimensions. It not only has themultiresolution and time-frequency localization properties, but also shows a very highdegree of directionality and anisotropy. Motivated by and capitalizing on this property,contourlet transform can be applied in a wide range of image processing tasks, such asdenoising, and has shown its potential in the field of medical image processing.The objectives of this dissertation were to introduce contourlet analysistechniques into the ultrasound image processing in HIFU system and make properadaption to clinical applications, including image statistical modeling, ultrasoundspeckle suppressing and treatment target edge detection for real-time HIFU therapy.Firstly, to further analyse the statistical characteristics of the contourletcoefficients for ultrasound image, we construct a new asymmetric piecewisegeneralized Gaussian function (APGGF) for image statistical modeling in transformdomain. By fitting the statistical probability more precisely and giving thecorresponding parametric estimation, this parametric modeling method has potentialfor many image processing applications, such as image coding, feature extraction,image denoising, and so on.Then we present a new contourlet-based speckle reduction method for medical ultrasound images. This method gives a scale-adaptive threshold in Bayesianframework based on modeling the subband contourlet coefficients of the ultrasoundimages after logarithmic transform as generalized Gaussian distribution. An adjustedproportional parameter is proposed for adapting the threshold to medical ultrasoundimages in contourlet domain. According to its less computing time, this ultrasoundimage pre-precessing algorithm can satisfy the HIFU intra-surgery requirement.For more precisely preprocessing the ultrasound images, a new specklesuppression method for medical ultrasound images based on contourlet transform wasproposed. Modeling the speckle with Rayleigh distribution in logarithmicallytransformed ultrasound images, a maximum a posterior (MAP) estimator is appliedfor speckle reduction via manipulating the coefficients in contourlet domain. Asgeneralized Nakagami distribution provides a better model for the statistic of speckledue to its capability and generality, we also proposed another speckle suppressionmethod according to generalized Nakagami distribution model, and further analysiswere then given for several special cases.A contourlet based edge detection method was then given to extracting theinside-body mark from ultrasound images for phased HIFU intra-surgery targetlocalization. This method extracts the curve structure in images by detecting modulusmaxima in different scale and different directional coutourlet subbands. Thecharacteristic of multiresolution and multidirection, and the low computationcomplexity make the contourlet based edge detection method a choice for phasedHIFU intra-surgery image processing.Finally, we discussed the affection of the respiration in HIFU therapy system andthe respiration control techniques in image guided radiotherapy. For HIFUintra-surgery target tracking, there are two key factors: the effectiveness ofinside-body mark detection in real time to locating the treatment target; the positionshift of the phased HIFU lesion is also fast enough to make tracking treatmentpossible. According to the characteristics of the image series scanning mode duringthe surgery system, a target localization method is then given in our clinical phasedHIFU therapy system.