Automatic Segmentation of Organs in Computed Tomography for Radiation Therapy Pla

描述(申请人提供)：医学视觉系统公司开发了一种新的图像分割技术，该技术提高了自动医学图像分割的准确性和稳健性。我们的商业目标是提供一种产品，以低成本产生准确、快速、全自动的分割，以辅助前列腺癌的调强放射治疗(IMRT)治疗。调强放射治疗(IMRT)的使用已经成为治疗前列腺癌的首选方法，其辐射剂量超过70GY[1]。这项技术的开发是为了确保治疗性剂量的辐射只照射到目标器官、前列腺和精囊，而不会影响附近的非受累结构，如膀胱和直肠。这项技术依赖于剂量规划软件和动态多叶准直器来保护这些敏感的器官。在没有预防措施的情况下，15%至35%的患者出现了2级或更严重的直肠毒性[2，19，20，22，31，32]，并且，较不确定的是，增加了晚期尿路并发症[5，12，30]。调强放射治疗计划软件需要精确分割腹部和骨盆的器官。包括邻近组织的前列腺分段可能会不必要地照射这些组织，而膀胱或直肠等包含太多周围组织的器官分段可能会干扰辐射剂量的完全传递。因此，不准确的分割对患者的生活质量有真正的影响。此外，手动分割感兴趣的器官所需的时间大大限制了可以进行的放射治疗的数量。此外，如果图像引导放射治疗(IGRT)以IMRT取代适形治疗的方式取代IMRT，那么人工分割将成为当代放射治疗中更有限的瓶颈。通过与三个癌症治疗中心建立伙伴关系，初步工作已经开始。研究人员在先前工作的基础上证明了带有AdaBoost的自动上下文模型(ACM)在磁共振(MR)成像中的皮质下分割是有效的[15，14，16，17，18]。我们建议在此基础上开发一种新的基于学习的图像分割系统，该系统能够准确地自动分割在不同成像设备上拍摄的腹部X射线计算机断层扫描(CT)中的感兴趣器官。腹部CT图像中器官的准确分割由于腹部解剖结构的巨大变化、腹部组织的运动、X射线计算机断层扫描中解剖结构之间的有限对比度以及成像设备、方案和技术的不同而变得复杂。在这份第一阶段的SBIR提案中，我们将解决现有分割工具的局限性，以实现调强放疗治疗计划所需的准确性和自动化。我们建议开发一种创新的基于学习的分割系统，使用自下而上和自上而下的方法。我们将构建一个新的特征词典，通过使用图像配准技术隐含地整合基于atlas的分割方法，并使用我们的合作者获取和分割的图像建立一个适度的“基本事实”数据库。第二阶段的提案将展示这些技术的临床可行性，解决系统在多个成像地点的稳定性，并通过提高分割准确性来确定拟议系统对患者预后的影响。 公共卫生相关性：医学视觉系统开发了一种新的基于学习的图像分割技术，可以同时提高自动医学图像分割的准确性和稳健性。我们的商业目标是提供一种以低成本产生准确、快速和全自动分割的产品，以辅助前列腺癌的调强放射治疗(IMRT)和未来的图像引导放射治疗(IGRT)。

DESCRIPTION (provided by applicant): Medical Vision Systems has developed a new image segmentation technology that improves the accuracy and robustness of automated medical image segmentation. Our commercial goal is to offer a product that produces accurate, fast, fully automatic segmentations at low cost to aid the Intensity Modulated Radiation Therapy (IMRT) treatment of prostate cancer. The use of Intensity-Modulated Radiation Therapy (IMRT) has become the preferred method for treating prostate cancer through radiation doses in excess of 70 Gy [1]. This technology has been developed to ensure that therapeutic doses of radiation are delivered only to the target organs, the prostate and the seminal vesicles, without affecting nearby non- involved structures, such as the bladder and the rectum. This technique relies on dose planning software and dynamic multileaf collimators to shield these sensitive organs. Without preventative measures, 15% to 35% of patients developed grade 2 or worse rectal toxicity [2, 19, 20, 22, 31, 32] and, with less statistical certainty, increased late urinary complications [5, 12, 30]. IMRT Treatment Planning software requires accurate segmentations of the organs in the abdomen and pelvis. Segmentations of the prostate that include neighboring tissues can irradiate those tissues unnecessarily, while segmentations of organs such as the bladder or rectum that include too much surrounding tissue can interfere with complete delivery of radiation dose. Thus, an inaccurate segmentation has a real effect on the quality of life of the patient. Furthermore, the time required to produce a manual segmentation of the organs of interest significantly limits the number of radiation therapy treatments that can be undertaken. In addition, if image-guided radiation therapy (IGRT) supplants IMRT in the same way that IMRT has supplanted conformal therapy, then manual segmentation will become an even more limiting bottleneck in contemporary radiation therapy. Preliminary work has begun through partnerships with three cancer treatment centers. The investigators have built upon previous work that shows the Auto Context Model (ACM) with AdaBoost is effective at subcortical segmentation in magnetic resonance (MR) imaging [15, 14, 16, 17, 18]. We propose to build upon this foundation to develop a new learning-based image segmentation system capable of accurately and automatically segmenting organs of interest in abdominal x-ray computed tomography (CT) scans taken at different imaging facilities. Accurate segmentation of organs in abdominal CT images is complicated by large variations in abdominal anatomy, motion of abdominal tissues, the limited contrast between anatomic structures in x-ray computed tomography, and variations in imaging equipment, protocols, and techniques. In this Phase I SBIR proposal, we will address the limitations of existing segmentation tools to achieve the accuracy and automation required for IMRT treatment planning. We propose to develop an innovative learning-based segmentation system using both bottom-up and top-down approaches. We will construct a novel feature dictionary, implicitly incorporate atlas-based segmentation methods through the use of image registration techniques, and build a modest "ground truth" database using images acquired and segmented by our collaborators. A Phase II proposal would demonstrate the clinical feasibility of these techniques, addressing both the stability of the system across multiple imaging sites and determining the impact of the proposed system on patient outcomes through increased segmentation accuracy. PUBLIC HEALTH RELEVANCE: Medical Vision Systems has developed a new learning-based image segmentation technology that can simultaneously improve the accuracy and robustness of automated medical image segmentation. Our commercial goal is to offer a product that produces accurate, fast, and fully automatic segmentations at low cost to aid in Intensity Modulated Radiation Therapy (IMRT) and, in the future, Image Guided Radiation Therapy (IGRT) for prostate cancer.