SCH: EXP: A Quantitative Platform for CT Colonography

计算机断层结肠镜检查（CTC）是指放射科医生在对准备好的患者进行腹部CT扫描后，对结肠表面（管腔）进行可视化检查。结肠息肉的切除是通过一种称为光学结肠镜检查的微创手术进行的。早期发现息肉的CTC和切除息肉的OC的适当同步是预防结肠癌最有效和最经济的方法，早期发现的治愈率超过97%。通过一系列图像分析步骤，可以构建结肠的三维（3D）表示，然后放射科医生使用虚拟摄像机将其可视化，以便检查结肠表面的异常和息肉。可视化包括提取三维冒号表示的中心线，这是虚拟摄像机的最佳基准。本课题提出了一个基于现代计算机视觉和图形学的CTC定量平台，从图像分辨率、成像伪影、结肠拓扑和息肉位置等方面对可视化过程进行量化和改进。在敏感性和特异性方面加强息肉的检测和分类将改善结肠癌的早期检测，结肠癌是死亡率和成本方面的主要国家医疗保健问题。除了计算机辅助制造和数字印刷之外，拟议的仿真平台还将应用于生物医学教育和培训，以及隐藏管状物体的表面检查和基于图像的可视化的工业应用。CTC的可视化包括四个连续步骤：i)过滤CT扫描以去除扫描仪噪声并解决部分体积效应；ii)对生成的图像进行分割，将结肠表面与腹部CT扫描中出现的腹部其他结构（如肝脏、胰腺和小肠）分离开来；iii)使用计算几何和图形生成3D冒号表示；iv)可视化的3D表示，由放射科医生使用虚拟摄像机检查腔面和检测结肠息肉。每一个步骤都是建立在坚实的数学基础上的，这些基础是在过去二十年中图像分析和计算机图形学文献中发展起来的。滤波可以使用各向异性扩散滤波来实现，而分割可以通过统计和变分方法的融合来实现，这种方法将结肠组织从其他解剖结构中分离出来，并提供分割结肠的连续/连接表示。分割冒号的三维重建是通过图形学中常用的方法来完成的，例如行进立方体算法。可视化是一个复杂的步骤，它包括三维重建中心线的生成和虚拟摄像机的合理分配。中心线可以通过变分演算和水平集法生成。可视化中常见的错误是由于成像的不确定性（例如，扫描仪引起的噪声，部分体积效应，预备残余和患者运动）和患者特定情况可能导致结肠重建扭曲或断开。这些错误可能会混淆息肉的位置、形状和质地，导致错误的诊断。本项目旨在开发一种新颖的CTC前端仿真，为虚拟摄像机提供精确的传感器规划。所提出的模拟将提供对CTC的独特理解，并将导致发现图像引导干预的方法，通过光学结肠镜检查和结肠切除术的手术计划有效地去除结肠息肉。该模拟平台将适用于生物医学教育、住院放射科医师和医疗保健专业人员的培训，并将促进生物医学领域以外的各种应用。

Computed-Tomography Colonoscopy (CTC) refers to visualization of the colon surface (lumen) by radiologists following an abdominal CT scan of prepped patients. Removal of colonic polyps is performed by a minimally-invasive procedure known as Optical Colonoscopy. A proper synchronization between CTC for early detection of polyps and OC for their removal is the most effective and economical approach to prevent colon cancer, which has over 97% rate of recovery with early detection. Through a sequence of image analysis steps, a three-dimensional (3D) representation of the colon can be constructed, which is then visualized by radiologists, using a virtual camera, in order to examine the colon surface for abnormalities and polyps. Visualization involves extraction of the centerline of the 3D colon representation, which is the optimum datum for the virtual camera. In this project, a quantitative platform for CTC, based on modern computer vision and graphics, is proposed to quantify and improve the visualization process with respect to image resolution, imaging artifacts, colon topology and polyps' locations. Enhancing the detection and classification of polyps in terms of sensitivity and specificity will improve early detection of colon cancer, a major national healthcare concern in terms of mortality and cost. The proposed simulation platform will have applications in biomedical education and training, and in industrial applications for surface inspection and image-based visualization of hidden tubular objects, in addition to computer-aided manufacturing and digital printing.Visualization in CTC consist of four sequential steps: i) filtering the CT scan for removal of scanner noise and resolving partial volume effects; ii) segmentation of the resulting images to isolate the colon surface from other structures in the abdomen, which appear in the abdominal CT scan (e.g., the liver, pancreas and small intestines); iii) generation of a 3D colon representation using computational geometry and graphics; and iv) visualization of the 3D representation, by radiologists, using virtual cameras to examine the lumen surface and detect colonic polyps. Each of these steps are based on solid mathematical foundation developed in the image analysis and computer graphics literature in the past two decades. Filtering may be achieved using anisotropic diffusion filtering, whereas segmentation can be achieved by fusion of statistical and variational methods, which separates the colon tissues from the other anatomies and provides a continuous/connected representation of the segmented colon. 3D reconstruction of the segmented colon is performed by common methodologies in graphics such the marching-cube algorithm. Visualization is an elaborate step, which involves generating the centerline of the 3D reconstruction, and a proper allocation of the virtual camera. The centerline may be generated by variational calculus and the level sets method. Common errors in visualization are due to the uncertainties in imaging (e.g., scanner-induced noise, partial volume effects, prep residuals and patient motion) and patient-specific circumstances which may lead to distorted or disconnected colon reconstructions. These errors may obscure the location, shape and texture of polyps, leading to erroneous diagnosis. This project aims at developing a novel simulation for the front-end of CTC, an accurate sensor planning for the virtual cameras. The proposed simulation will provide unique understanding of CTC and will lead to discovery of methods for image-guided interventions to efficiently remove colonic polyps by Optical Colonoscopy and surgical planning for colectomy. The simulation platform will be applicable in biomedical education, training of resident radiologists and healthcare professionals, and will also promote various applications outside the biomedical domain.