TR&D Project 2: Virtual Scanners

摘要-TRD2：虚拟扫描仪 虚拟成像试验(VITs)为开展计算机断层扫描(CT)研究提供了一种强有力的替代方案 以人类为研究对象的技术。由于试验在Silico中进行，虚拟试验需要快速和现实的 CT模拟器。然而，由于代表性有限，现有的CT模拟器不能满足这一需求 实际的CT采集过程速度慢。使用蒙特卡罗方法的模拟器在以下方面是最佳的 对图像采集过程进行精确建模，但对于模拟高分辨率图像来说太慢。 替代的射线跟踪方法速度更快，但无法提供实际的吸收辐射剂量估计， CT成像中的一个非常重要的因素。大多数模拟器在建立特定模型的能力上受到进一步的限制 CT制作和模型，这将是代表实际的临床CT成像场景所必需的。 该项目开发并提供了一个新的CT仿真平台，以满足预期的吞吐量和真实感 虚拟成像试验。该平台结合了高空间/时间细节的优点(由射线提供- 跟踪)、精确的辐射剂量和散射估计(由蒙特卡洛提供)、速度(由GPU提供 计算和熟练编程)、专用性(基于精密系统的CT子组件建模 CT制造商提供的规格)。已经为一台CT扫描仪制作了原型，该项目将扩展 为多个CT系统搭建了一个综合的CT模拟器平台。 该项目的具体目标是(1)对CT采集子组件进行详细建模；(2)对CT进行建模 用于估计初级信号、散射和辐射剂量的采集方案；(3)实施以下过程 集成、图像形成和验证；以及(4)构建模块化接口，以便能够有效地使用 模拟器。模拟将包括制造商特定的、用户定义的和通用的(即，制造商- 中性)CT系统和重建算法、探测器几何结构和模型(包括光子计数 探测器)、对采集规格的完全用户控制(即，来自TRD1的虚拟患者输入，CT扫描仪， 协议、千伏、毫安、侦察等)，以及带有图形用户界面和基于脚本的实用程序的用户友好的模块化界面。 这项工作将提供首个快速而准确的CT模拟器，具有扫描仪专用、用户- 可定制的通用3D和4D建模功能，可以模拟这两种重建图像 和吸收辐射剂量。用户将能够利用模拟器学习各种CT技术 和应用，例如与辐射剂量优化、图像质量评估和 心脏和呼吸运动引起的图像变形。模拟器将实现基于任务的设计和 评估新的CT系统和基于人工智能(AI)的培训，通过生成大规模 逼真的图像数据集，复制了临床图像的真实感，并增加了已知基础的优势 真心话。CT仿真平台，结合虚拟患者套件(TRD1)和虚拟阅读器(TRD3) 由该中心提供，构成了在CT中启用虚拟成像试验的基本工具集。

ABSTRACT – TRD2: Virtual Scanners Virtual Imaging Trials (VITs) offer a powerful alternative to conducting studies of computed tomography (CT) technologies with human subjects. With the trial taking place in silico, virtual trials require a fast and realistic CT simulator. However, current CT simulators are inadequate to meet this need due to limited representation of the actual CT acquisition processes and slow speed. Simulators using Monte Carlo methods are optimal in accurately modeling the image acquisition process but too slow for simulating high resolution images. Alternative ray-tracing methods are faster but unable to provide realistic estimates of absorbed radiation dose, a factor of high importance in CT imaging. Most simulators are further limited in their ability to model specific CT makes and models, which would be essential to represent an actual clinical CT imaging scenario. This project develops and provides a new CT simulation platform to meet the desired throughput and realism of virtual imaging trials. The platform combines the benefits of high spatio/temporal details (provided by ray- tracing), precise radiation dose and scatter estimates (provided by Monte Carlo), speed (provided by GPU computing and proficient programing), and specificity (modeling CT subcomponents based on precise system specifications from CT manufacturers). Already prototyped for one CT scanner, this project will expand the prototype into a comprehensive CT simulator platform for multiple CT systems. The Specific Aims of the project are (1) to model CT acquisition subcomponents in detail; (2) to model CT acquisition schemes for estimating primary signal, scatter, and radiation dose; (3) to implement processes for integration, image formation, and validation; and (4) to build a modular interface to enable effective use of the simulator. The simulation will include manufacturer-specific, user-defined, and generic (i.e., manufacturer- neutral) CT systems and reconstruction algorithms, detector geometry and models (including photon-counting detectors), full user-control over acquisition specifications (i.e., virtual patient input from TRD1, CT scanner, protocol, kV, mA, recon, etc.), and a user-friendly modular interface with both GUI and script-based utility. This work will provide a first-of-its-kind rapid and accurate CT simulator with scanner-specific, user- customizable, and generic 3D and 4D modeling capabilities, which can simulate both reconstructed images and absorbed radiation dose. Users will be able to utilize the simulator to study a variety of CT technologies and applications, such as those pertaining to radiation dose optimization, image quality assessment, and image deformation from cardiac and respiratory motion. The simulator would enable task-based design and evaluation of new CT systems and artificial intelligence (AI)-based training through generating large-scale realistic image datasets that replicate the realism of clinical images with the added advantage of known ground truth. The CT simulation platform, combined with the suite of virtual patients (TRD1) and virtual readers (TRD3) offered by the Center, form the essential toolset to enable virtual imaging trials in CT.