六边形多源CT结构及混合投影分离算法研究

In a conventional CT system, the reconstructed object is often traversed by single X-ray source at each time. However, the conventional CT system exists some drawbacks. Firstly, almost all CT scanners collect the projection sequentially by adopting a step-and-shoot process, which demands high x-ray power and high speed of gantry rotation. Secondly, the speed of the collcted projction data depends on the mechanical rotation speed of CT rotation gantry. Finally, the "bad" projection data often arise from the mechanical rotation of the gantry, which will result in the artifacts of the reconstructed object. To overcome the aforementioned drawbacks, the simultaneous collection structure of projection named "Hexagonal Imaging Structure of the Multi-Source CT" is proposed in this project, which is based on the carbon nanotube x-ray array. Based on the proposed hexagonal imaging structure, the coding matrix generated from the mean square error criterion indicates that the sources are off or on at a given time. In this case,the coding matrix is exploited to decompose the mixing projection data, which are collected from the multisource CT detector. By using the proposed method, a drastic increase of the speed and reduction of the x-ray peak power can be achieved without compromising the imaging quality.

在常规CT系统中，每次只能允许一个X射线源照射待重建物体，由探测器采集投影数据，然后X射线源再转到下个位置再重复原来的方式采集投影数据，在这种情况下存在几个问题，首先:CT投影数据的采集需要极快的机架旋转速度和极高的X射线峰值功率;其次：投影数据的采集速度受限于X射线源的机械转动速度,而且X射线源的机械转动的到达的位置并不十分精确，容易造成重建过程中出现伪像。鉴于此，本文拟提出一个基于碳纳米管新的数据并行采集结构，即"六边形多源CT成像结构"，加快数据采集速度，来避免机械转动所带来伪影问题。在新的模式下，采用本文拟提出的基于MSE准则的编码矩阵算法来控制多个射线源的通断，之后同时照射待重建物体来采集混合投影数据，并依据所采用的编码矩阵对混合投影进行分离，从而可以保证每个射线源都有对应的投影值。在保证投影数据质量前提下，提高了数据采集速度和降低X射线峰值功率，并避免了X射线管和探测器的转动。

在六边形多源CT结构及混合投影分离算法研究与应用中共发表SCI检索论文4篇，其中IEEE 汇刊文1篇。项目总体主要研究成果：1）提出了扇束DPC-CT的近似重建方法。该方法的基本思想是根据物体临近断层之间的相关性，首先将螺旋扫描条件下采集到的数据插值到某个断层上，估计出该断层的正弦图，再根据“正反投影法”提取折射角，最后通过扇形束重建算法完成重建。2）提出了基于潜在语义的图像盲去模糊的分离算法，提高了图像去模糊的效果。该方法采用矩阵分解方法提取图像的潜在语义，并建立起潜在语义与待恢复图像之间的约束关系，从而提高了恢复效果。3）提出了基于结构稀疏重构的哈希函数学习方法。通过对相似的点用L21范数对重构系数进行稀疏约束以增强局部保持映射过程中的判别性，并在调和协方差矩阵和最小化数据的重构误差间建立了一种平衡机制。4）提出了基于改进字典学习的异常检测方法。该方法从原始训练数据中学习典型事件类型，同时挖掘样本空间的分布信息。由于在训练数据中挖掘出了典型的事件类型，因此在同源的测试数据中正常的样本被更好的表达，增大了异常样本与正常样本编码间的差异，使得异常的样本更容易区分。同时探索了样本间的相似性，使得相似样本间的编码也相似，提高了编码的可靠性。