Research On Edge Contour Extraction Method In Tumor Therapy Using Focused Ultrasound

As a medical therapy technology for tumors with extracorporeal ablation,the high intensity focused ultrasound(HIFU)ablation has many advantages compared to the traditional surgical treatments,such as non-invasion,high treatment effect,speedy recovery and shorter hospitalization.So this technology is widely used in clinical treatment of tumors in the past decades.In the process of the HIFU surgery,the ultrasound image surveillance system plays an important role to offer the operation information for doctors.However,the surveillance system just monitor the surgical procedure.The doctor need to identify the tumor location and manually delineate the tumor boundaries based on their eyes and experiences,and then use the ultrasonic probe to ablate the tumor tissue point by point.Because of the limitations with the manual operation,the HIFU ablation is very subjective.So the HIFU image-guided system is introduced to solve this problem,which is the improvement of the HIFU surveillance system.The basic idea of the HIFU image-guided system is to use the technology of the digital image processing instead of the manual delineation.After using the suitable algorithms,the computer can get the location and the boundaries of the tumors in ultrasound images precisely,and then provides the visual information for doctors to complete the surgical procedure.But due to limitations of imaging mechanism,the ultrasound images often have serious speckle noises and low contrasts,which lead to the blur in the target tumor boundaries.It is very intractable for the tumor boundary detection in HIFU image-guided system.So the research of this paper focuses on the de-noising methods and the segmentation methods for the HIFU ultrasound images in order to offer precise boundary information for the HIFU image-guided system.After studying the filtering algorithms based on the anisotropic diffusion and the image segmentation algorithms based on the level set methods and the active contour model,we proposed several improved algorithms,which include the following aspects:(1)For the de-noising of the HIFU ultrasound images,an improved SRAD model is proposed.The model redefine the diffusion coefficient of the traditional SRAD model with hyperbolic tangent function.After the re-construction,the improved SRAD model eliminates the block effect in the uniform region of the images,and then a damping factor is used to guide the speed of attenuation in the no,uniform area.So it can reserve the details and the weak edges of the images.In addition,the median absolute deviation is used to estimate the initial scale for the speckle noise scaling function,which avoids the uncertainty of the manual setting.At last,to monitor the degree of the filtering process automatically,the relative smooth increment is introduced.The iteration process of the partial differential equation can stop adaptively,which avoid the filtering instability caused by the manual setting.The experiments show out the proposed method can not only filter the speckle noises of the images effectively,but also eliminate the block effect problem caused by the traditional anisotropic diffusion,and it can also improve the ability of preserving the detail information of the images.So the filtering performances of the proposed de-noising algorithm are better than the traditional ones.(2)On the basis of studying the region-based active contour model,according to the merits and demerits of the global region active contour model and the local region active contour model,a hybrid active contour model based on global and local fitting energy is proposed.After using the difference image of the original image to construct the global fitting energy,the model can not only keep the global performance,but also introduces the edge information of the image in the energy equation,and then combined with the region-scalable fitting energy to constitute for the energy functional.The experiments show out the hybrid active contour model inherits the advantages of the both fitting energies and avoids the disadvantages of them,which has satisfying segmentation results while dealing ultrasound images with intensity inhomogeneity.(3)For the HIFU ultrasound image segmentation,aiming at eliminate the influence of the spurious boundaries in the ultrasound images,an oriented distance regularized level set evolution model(ODRLSE model)based on the spurious boundaries suppression is proposed.The traditional DRLSE model only considers the gradient amplitude of the images,and neglects the gradient direction,so it can’t distinguish the target boundaries and the spurious boundaries.According to the direction relations between the normal vector of the initial contour and the gradient vector field,the improved DRLSE model reconstruct the edge map of the original images,and use the new edge map to construct the edge detection function of the traditional DRLSE model.The experiments show out the improved ODRLSE model can suppress the influence of the spurious boundaries constituted by the normal tissues in human bodies and segment the target boundaries precisely.(4)For the HIFU ultrasound image segmentation,in order to deal with the weak edge problems in the ultrasound images,an improved level set evolution model based on the multi-scale filter(MS-DRLSE model)is proposed.On the basis of a certain interval,several Gaussian standard deviations of the edge detection function are used as scale images,several scale images are set up,and along the scale’s descent direction,the above gradient vector field in each pixel is fixed according to the direction relations of the gradient vector field between the adjacent scales.The proposed multi-scale filter method can filter the noise effectively and enhance the target boundaries as well.Meanwhile,to solve the one-way evolution problem of the balloon force,the Laplace operator of images after the multi-scale filter is used to replace the original balloon force.So the model can not only realize the two-way evolution,but also improve the ability of the weak edges detection in the images.The experiments show out the improve level set evolution model based on the multi-scale filter can have a good segmentation result while dealing with the weak edges and the edges with complex structure in the ultrasound images.