数字乳腺体层合成成像关键技术研究

发布时间：2018-05-11 21:19

本文选题：数字乳腺体层合成 + 乳腺图像对比度增强　；参考：《南方医科大学》2017年博士论文

【摘要】：乳腺癌已成为女性的头号癌症,其防控关键在于普及乳腺筛查。数字乳腺X射线摄影(Digital Mammography,DM)是乳腺筛查首选的影像检查手段,其具有空间分辨率高、辐射剂量低等优点,但无法解决三维组织堆叠的问题。数字乳腺体层合成(Digital Breast Tomosynthesis,DBT)是基于DM发展的新成像技术,通过射线源进行有限角度低剂量曝光,采集有限数量的投影并进行近似三维重建,所得体层图像能有效解决DM的组织重叠问题并提供更丰富的诊断信息。国内DBT系统及成像方法的研究较少,本文针对乳腺成像特点和体层合成技术进行乳腺2D/3D成像关键技术研究,主要工作如下:(1)提出并实现一种基于LP-DUM乳腺图像多尺度对比度增强算法。针对乳腺2D图像差异小等问题,在LP多尺度分解中,利用乳腺局部调整项和全局约束项构建可变权重并进行动态非锐化掩膜增强,以实现局部去噪和边缘增强。通过体模和乳腺增强实验可得,本文算法结果对比度增强效果好且无明显噪声放大,信息熵和细节对比更丰富,较好保留轮廓和边缘同时提升图像观感。(2)提出并实现一种基于SATpV的DBT快速迭代重建算法。针对DBT有限角度低剂量欠采样问题,以SART为基础,提出前向和反向投影的权重快速计算方法,并针对图形处理单元进行并行重建架构优化,同时提出选择性各向异性全变差项(SATpV)进行正则化约束,最后利用投影重排策略加快重建收敛速度。通过仿真体模和样机数据进行对比实验,结果证明本算法的重建效果最佳且无明显失真,9秒内完成五次迭代重建,满足临床应用时效需求。(3)提出并实现一种基于图像质量驱动的几何参数标定研究。针对机架装配和抖动等误差导致重建图像存在几何伪影此问题,利用解析法进行初始几何参数集求解,再以初始解进行图像重建,对所得图像的梯度范数作图像质量并以此作为最优化目标函数进行寻优,最终获得几何参数集最优解。通过仿真和样机数据进行验证,结果显示本算法所得结果在CBCT/DBT图像重建质量改善显著,几何标定精度较高。(4)提出并实现一种基于全局投影修正的截断伪影抑制及图像重建算法。针对有限尺寸探测器和大角度扫描造成投影截断而产生阶梯截断伪影此问题,在DBT图像迭代重建中同步计算相邻投影截断差异容积,利用全局投影通过WBP修正对应容积图像,并利用邻域背景基值补偿截断容积的体素,依次进行前向和逆向投影更新重建。通过仿真和真实体模进行对比测试,结果表明本方法所得图像无明显截断伪影存在,伪影抑制效果明显。本文针在图像增强、图像重建、几何校正和截断伪影抑制等关键技术均取得初步成果,但由于实验条件和时间等限制,相关研究有待深入探究。
[Abstract]:Breast cancer has become the number one cancer in women, the key to prevention and control lies in universal breast screening. Digital mammography (Digital mammography) is the preferred method for mammary gland screening. It has the advantages of high spatial resolution and low radiation dose, but it can not solve the problem of three-dimensional tissue stacking. Digital Breast Tomosynthesis (DBT) is a new imaging technology based on DM, which uses a limited angle low dose exposure, a limited number of projections and approximately 3D reconstruction. The resulting images can effectively solve the tissue overlap problem of DM and provide more abundant diagnostic information. There are few studies on DBT system and imaging methods in China. This paper studies the key techniques of breast 2D/3D imaging according to the characteristics of mammography and the technology of body layer synthesis. The main work is as follows: 1) A multi-scale contrast enhancement algorithm based on LP-DUM breast image is proposed and implemented. To solve the problem of small difference in 2D mammary gland image, variable weight is constructed by using local adjustment item and global constraint item in LP multiscale decomposition, and dynamic unsharp mask enhancement is carried out to realize local denoising and edge enhancement. Through the experiments of phantom and mammary gland enhancement, the results of this paper show that the contrast enhancement effect is good, no obvious noise is amplified, and the information entropy and detail contrast are more abundant. A fast iterative reconstruction algorithm for DBT based on SATpV is proposed and implemented. Aiming at the problem of DBT limited angle low dose under-sampling, based on SART, a fast weight calculation method based on forward and backward projection is proposed, and the parallel reconstruction architecture is optimized for the graphics processing unit. At the same time, the selective anisotropic total variation term (SATpV) is regularized, and the projection rearrangement strategy is used to speed up the reconstruction convergence. The results show that the reconstruction effect of this algorithm is the best and has no obvious distortion, and five iterations are completed in 9 seconds. A geometric parameter calibration study based on image quality driving is proposed and realized. In order to solve the problem of geometric artifacts caused by the errors of frame assembly and jitter, the initial geometric parameter set is solved by analytic method, and then the initial solution is used to reconstruct the image. The gradient norm of the obtained image is used as the image quality and is used as the optimization objective function to obtain the optimal solution of the geometric parameter set. The results of simulation and prototype data show that the proposed algorithm improves the quality of CBCT/DBT image reconstruction significantly, and the geometric calibration accuracy is high. 4) A truncated artifact suppression and image reconstruction algorithm based on global projection correction is proposed and implemented. In order to solve the problem of truncation artifacts caused by projection truncation caused by finite size detectors and large angle scanning, the differential volume of adjacent projection truncation is calculated synchronously in the iterative reconstruction of DBT images, and the corresponding volume images are corrected by global projection through WBP. The neighborhood background basis is used to compensate the volume of voxels in truncated volume, and then the forward projection and reverse projection are used to reconstruct the voxels in turn. The results show that there is no significant truncation artifact in the images obtained by this method, and the effect of artifact suppression is obvious. The key techniques of image enhancement, image reconstruction, geometric correction and truncation artifact suppression have been achieved in this paper. However, due to the limitations of experimental conditions and time, the relevant research needs to be further explored.
【学位授予单位】：南方医科大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：R737.9;TP391.41

【相似文献】