Development of a low-cost hardware accelerator for 3D image registration.

描述（由申请人提供）：我们的总体目标是开发一种新型计算解决方案，用于在1分钟内自动准确配准（空间对齐）任何模态和任何解剖结构（刚性或可变形）的三维（3D）医学图像。这种能力目前还不存在。现有的图像配准解决方案具有有限的准确性和/或有限的适用性，阻止了广泛和常规的临床使用。基于先前的重要学术研究，我们将在这个为期1年的STTR第一阶段项目中证明创建拟议技术的可行性。在证明可行性之后，我们将创建硬件加速图像配准的全功能原型}一个紧凑的、相对低成本（制造成本：~$20，000）的PC板}，以商业化为最终目标。图像配准是现代医学中的基本需求，这是一种仍未得到满足的需求。这是在融合具有互补信息的图像之前必要的第一步，或者可以减去在不同时间拍摄的图像以量化解剖/生理变化。图像配准具有许多其他应用，包括在许多新兴的微创图像引导干预中的术前和术中图像的配准。通过系统级仿真和性能表征（我们的两个具体目标），我们将在第一阶段创建所提议的硬件加速器的软件仿真。我们将使用这款功能相同、速度相似的模拟器配准大脑、肺部和腹部的现有图像，以在1分钟或更短时间内显示准确的图像配准。满足这些性能里程碑将提供证据证明可以创建这种临床可行的多用途计算解决方案，并证明过渡到第II阶段和继续资助是合理的。我们提出的低成本、超快、易于使用和准确的计算解决方案有望释放医学图像配准的全部潜力。项目健康相关性：结合2个或更多不同类型的医学图像，可以提供有关患者状况的更准确信息。比较不同时间拍摄的图像有助于监测疾病对治疗的反应。在任何一种情况下，图像配准（对齐）都是关键的第一步。目前的图像配准方法速度慢、复杂、繁琐，实用性有限。我们建议开发适用于大多数器官和图像类型的自动，高速，三维配准功能。我们会先在第一期研究这项技术的可行性，然后才在第二期全面发展这项技术。}

DESCRIPTION (provided by applicant): Our overall goal is to develop a novel computing solution for automatic and accurate registration (spatial alignment) of three-dimensional (3D) medical images of any modality and any anatomy (rigid or deformable) in 1 minute. Such capability currently does not exist. Existing image registration solutions have limited accuracy and/or limited applicability, preventing wide and routine clinical use. Building on significant prior academic research, we will demonstrate the feasibility of creating the proposed technology in this 1-year STTR Phase I project. After demonstrating feasibility, we will create a fully functional prototype}a compact, relatively low- cost (manufacturing cost: ~$20,000) PC board}of hardware-accelerated image registration, with commercialization as the ultimate goal. Image registration is a fundamental need in modern medicine}a need that remains unmet. It is the necessary first step before images with complementary information can be fused, or images taken at different times can be subtracted to quantify anatomic/physiologic changes. Image registration has numerous other applications, including the registration of pre- and intraoperative images in a host of emerging minimally invasive image-guided interventions. Through system-level simulation and performance characterization (our 2 specific aims), we will create a software simulation of the proposed hardware accelerator in Phase I. We will register existing images of the brain, lungs, and abdomen using this functionally identical and similar-speed simulator to show accurate image registration in 1 min or less. Meeting these performance milestones will provide evidence that this clinically viable and multipurpose computing solution can be created and will justify transitioning to Phase II and continued funding. Our proposed low-cost, ultrafast, easy-to-use, and accurate computing solution promises to unlock the full potential of medical image registration. PROJECT HEALTH RELEVANCE: Combining 2 or more medical images of different types gives more precise information on a patient's condition. Comparing images taken at different times helps monitor how a disease is responding to treatment. In either case, image registration (alignment) is the crucial first step. Current image registration methods are slow, complex, and tedious, with limited practical applicability. We propose developing automatic, high-speed, three-dimensional registration capabilities that are applicable to most organs and image types. We will demonstrate the feasibility of creating such a technology in Phase I before its full development planned for Phase II. }