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

As a non-invasive medical therapy for tumors, High intensity focused ultrasound (HIFU) ablation has many virtues such as no wound, minor pain, high treatment efficacy and fast postoperative convalescence, and has been widely used in clinical treatment over the past few decades. Ultrasound image monitor plays an important role in HIFU ablation, because it provides a visual surgical environment for doctors. However, the existing ultrasound image monitor of HIFU system is mainly used for video surveillance. During the HIFU ablation, the tumor locations are manual delineated by doctors based on their experience. Because of the limitation of the manual operation, HIFU ablation has poor location precision, low efficiency and long operation time. In order to implement HIFU ablation more safely, efficiently and precisely, ultrasound image-guided technology is employed in HIFU system, which is one of the most important developing directions for HIFU system. Based on the ultrasound image processing technique, the computer could automatically identify the target tumor and obtain the tumor's location. Then the computer provides visual information for doctors to implement HIFU ablation in target area. However, because of the characteristics of ultrasound imaging, ultrasound images are always in low contrast and the edges of the target tumors are blurred, which is detrimental for tumor localization and boundary detection in HIFU system. Therefore, this paper researches on the boundary detection method in HIFU ultrasound imagery to provide precise locations of the target tumors for HIFU ultrasound image-guided system. This study covers two aspects including the method for HIFU ultrasound image filter and the method for image segmentation. The speckle noise suppression algorithm based on anisotropic diffusion filter and the ultrasound image segmentation algorithms based on active contour model are introduced in detail. In addition, the improved methods are presented, including the following several aspects:(1) For noise reduction of HIFU ultrasound images, an 8-neighbourhood SRAD model without parameter settings is proposed. In the improved model,4-neighbourhood is replaced with 8-neighbourhood to utilize the diagonal information, and the median absolute deviation is employed to effectively determine the initial threshold scale for the speckle scale function. In addition, the relative smooth index increment (RSII) is defined to stop the iterations adaptively. The filtering results show that the improved SRAD model is far superior to those traditional models in speckle noise reducing and tumor boundary preserving. Because the proposed model has no artificial operation in parameter settings and could adaptively stop the iterations, the uncertainties of the filtering results from artificial operations are greatly reduced, which provides favorable conditions for HIFU ultrasound image segmentation.(2) For HIFU ultrasound image segmentation, an improved gradient and direction vector flow model is proposed to deal with weak boundary segmentation in HIFU ultrasound imagery. In order to solve the problems that the traditional G&DVF snake model encountered in practical ultrasound image segmentation, the small directional line segments are replaced with fold line segments and different segments are assigned with different weight coefficients in the improved model. And a novel vector field that integrates the prior direction information from the directional segments with the conventional gradient vector field is defined to replace the conventional gradient vector field to diffuse and finally generate the vector field of the external G&DVF force. The experimental results show that the improved G&DVF snake model can obtain good segmentation results in ultrasound images with non-uniform intensity and complex noise, and has great performance in segmenting weak boundaries and long, thin boundary indentations.(3) For HIFU ultrasound image segmentation, a multi-scale gradient vector flow (MS-GVF) snake model is proposed to segment relatively clear tumor boundary based on multi-scale filter and spurious boundaries suppression. The Gaussian standard deviation is assigned as a scale in the proposed model. And the HIFU ultrasound images are segmented in different scales along the scale's descent direction, which is helpful in extending the capture range of the external force, avoiding the snake points being attracted by large noise points and keeping the detailed information of the target boundary. In order to reduce, even remove the negative effect of spurious boundaries in GVF vector field, the gradient directional information and the distance magnitude information of a newly defined distance map are used to suppress the spurious boundaries in the last scale. In addition, in the MS-GVF snake model, the GVF energy functional is expressed as an augmented Lagrangian function, and the FFT algorithm is used to calculate the GVF vector field, which could effectively improve the speed of the algorithm. The experimental results show that the MS-GVF snake model is accurate and robust, and the parameter settings are simple and easy to operation.(4) The applications of the improved G&DVF snake model and the MS-GVF snake model in HIFU ultrasound image segmentation are investigated. The improved G&DVF snake model is suitable for segmenting weak boundaries and long, thin boundary indentations. However, the weighting coefficient and the location setting of each fold line segment in G&DVF snake model have great effect on the final segmentation result. In the MS-GVF snake model, the capture range of the external force is greatly extended, the negative effect of spurious boundaries in image segmentation is effectively reduced, and the detailed information of the target boundary is reserved. The MS-GVF snake model is accurate in ultrasound image segmentation with simple parameter setting, while its performance in weak boundary segmentation is unsatisfied. Based on comparison between the advantages and disadvantages of these two models, this paper suggests that both the improved G&DVF snake model and the MS-GVF snake model are adopted to segment HIFU ultrasound images. The MS-GVF snake model is used to segment relatively clear tumor boundary, and the improved G&DVF snake model is used to segment weak boundaries.