Accurate Localization of General Tubular Structures in Medical Images

描述（由申请人提供）：许多医疗应用需要或将受益于医学图像中管状结构的自动定位，例如神经，血管等。近年来，一种特殊的算法在耳内面神经和鼓室索两种神经的定位上取得了较好的效果。该算法的结构具有开放性，可以解决当前许多定位方法存在的一些严重缺陷。因此，本项目的具体目的是：(1)进一步验证该算法在面神经和脊索上的性能。该方法已在同一台扫描仪上的14个CT上进行了测试，所有病例的结果都非常积极。该方法将通过在四家医院的60台CT上测试算法的准确性来进一步验证，这些CT将从目前NIH资助的项目中获得。(2)开发自动训练算法的软件。对于要定位的每种类型的结构，算法必须训练一次。目前，训练过程是手动的，非常耗时，并且需要先进的图像处理专业知识。这一目标将通过广泛的数学和图像处理文献综述、对该领域当前方法的综合误差分析以及创造性实验来实现。成功将通过定位结果的可接受性和定位的准确性与通过手动调优实现的定位相比来衡量。(3)将技术拓展到新的应用领域。乙状窦、颈动脉、视神经、视交叉和视束的定位结果将被检查。磁共振和CT将从NIH资助的3个项目中获得。结果的解释将由每个结构相关领域的经验丰富的医生协助。本研究中包含的解剖结构与三个应用相关。研究人员验证了一种用于放置人工耳蜗的新型微创技术，他们发现，当本研究中的四个结构被定位并呈现为3D时，新技术变得更加容易和安全。放置深部脑刺激器的医生发现视神经束的定位可以增强电极的放置。用放射疗法治疗头部癌症的医生需要准确定位视神经、束和交叉，以确保安全免受有害影响。

DESCRIPTION (provided by applicant): Many medical applications require or would benefit from automatic localization of tubular structures in medical images, e.g. nerves, vessels, etc. Recently, a particular algorithm was developed which has performed robustly in localizing two nerves in the ear, the facial nerve and chorda tympani. The structure of this algorithm is open to additional designs which could address some severe shortcomings in many current localization methods. The specific aims of this project are thus to:(1) Further validate the performance of this algorithm on the facial nerve and chorda. The method has been tested on 14 CT's, taken from the same scanner, with very positive results in all cases. The method will be further validated by testing the accuracy of the algorithm on 60 CT's from four hospitals which will be made available from a current NIH funded project. (2) Develop software which automatically trains the algorithm. The algorithm must be trained once for each type of structure it is to localize. Currently, the training process is manual, very time consuming, and requires advanced image processing expertise. This aim will be accomplished through extensive mathematical and image processing literature review, comprehensive error analysis of current methods in the field, and creative experimentation. Success will be gauged by acceptability of localization results and accuracy of localizations compared to those achieved through manual tuning. (3) Extend the technology to new applications. Localization results for the sigmoid sinus, carotid artery, optic nerves, optic chiasm, and the optic tracts will be examined. MR's and CT's will be acquired from 3 NIH funded projects. Interpretation of results will be assisted by experienced physicians in the field for which each structure is relevant. Anatomical structures included in this study are relevant to three applications. Researchers validating a new minimally invasive technique for placing cochlear implants have found that the new technique is much easier and safer when four structures in this study are localized and rendered in 3D. Physicians placing deep brain stimulators have found that localization of the optic tracts would enhance electrode placement. Physicians treating head cancers with radiotherapy require accurate localization of the optic nerves, tracts, and chiasm to ensure safety from harmful effects. PULIC HEALTH RELEVANCE: The results of this project have the potential to provide physicians with anatomical information that will make cochlear implantation safer and a more available treatment for deafness, increase safety and efficiency in the placement of deep brain stimulators for treatment of Parkinson's, increase safety in the treatment of cancer by guided radiotherapy, and benefit many other applications not addressed in this study.