Quantitative Image Modeling for Brain Tumor Analysis and Tracking

 描述（由申请人提供）：尽管我们对脑癌的生物学、诊断和治疗的认识有了巨大的进步，但在过去的三四十年里，与脑癌和其他中枢神经系统（CNS）癌症相关的死亡率一直保持稳定。早期诊断和治疗的进一步进展可能部分与改进普遍用于分析和分割脑肿瘤的计算模型有关。临床应用仍然需要改进难以检测的增强差、多病灶和小肿瘤的分割，这些肿瘤被多个异常组织（如水肿、坏死和囊肿）包围。此外，需要改进计算模型，以处理不同组织类型之间的扩散边界，从而实现稳健的脑肿瘤分割（BTS）。此外，为了减少认知后遗症，当代方案采用风险适应性治疗，其中风险分层是基于手术切除后残留肿瘤的体积和诊断时转移性疾病的存在。因此，如果不改善肿瘤体积的定量，就不可能进一步改善癌症结局，特别是儿童。此外，在不同的成像中心、研究、患者群体（成人和儿童）和设备之间复制先进的计算机算法是整个计算医学成像领域的一个持续问题。 因此，该研究项目的总体假设是，可以开发一种强大的自动BTS和其他异常和正常脑组织分割，用于定量和跟踪肿瘤体积，这反过来将有助于改善CNS肿瘤的早期诊断、随访和治疗。所提出的项目旨在专注于使用大量的BTS神经成像数据集的原则性计算建模，这些数据集正在变得普遍，特别是来自国家癌症研究所赞助的脑肿瘤分割（BRATS）挑战（http：www.braintumorsegmentation.org）。这一目标将通过以下目标实现：（1）使用来自不同成像中心的多模态MRI识别新特征、多类（组织）特征选择和难以检测的肿瘤和相关异常的分割;（2）通过融合基于图谱的肿瘤分割（ABTS）和基于特征的BTS（FBTS），实现肿瘤、其他异常和正常组织的稳健分割以及脑肿瘤的定位;（3）将软件集成到广泛可用的工具中（3D切片器）可通过多个NIH赞助的资源中心获得，例如神经影像分析中心（NAC），国家医学图像计算联盟（NA-MIC）和国家图像引导治疗中心（NCIGT），以更广泛地传播BTS工具;以及（4）验证和评估我们的综合BTS工具，以量化可检测性、灵敏度和特异性以及相应误差的改进。

 DESCRIPTION (provided by applicant):Mortality rates related to brain and other Central Nervous System (CNS) cancers have held steady over the last three or four decades, despite tremendous advancements in our knowledge about the biology, diagnosis, and treatment of brain cancer. Further progress in early diagnosis and treatment is likely to be associated, in part, with improving computational models that are used ubiquitously for analyzing and segmenting brain tumors. Clinical applications continue to necessitate improved segmentation of hard-to- detect poorly enhanced, multi-foci and small tumors that are surrounded by multiple abnormal tissues such as edema, necrosis and cysts. In addition, computational models need to be improved for handling diffusive boundaries among different tissue types for robust Brain Tumor Segmentation (BTS). Furthermore, in an effort to reduce cognitive sequelae, contemporary protocols employ risk-adapted therapy in which risk stratification is based on the volume of residual tumor after surgical resection and the presence of metastatic disease at diagnosis. Therefore, further improvement in cancer outcomes, particularly among children, is unlikely to be achieved without improved quantitation of tumor volume. Furthermore, replicating advanced computer algorithms across different imaging centers, studies, patient populations (adult and pediatric) and equipment is a persistent problem for the entire field of computational medical imaging. Consequently, the overall hypothesis of this proposed research project is that a robust automatic BTS and other abnormal and normal brain tissue segmentation can be developed for quantitation and tracking of tumor volume which, in turn, will help improve early diagnosis, follow- up and treatment of CNS tumors. The proposed project aims to focus on principled computational modeling using a huge amount of neuroimaging datasets for BTS that are becoming prevalent, especially from the National Cancer Institute's sponsored Brain Tumor Segmentation (BRATS) challenges (http://www.braintumorsegmentation.org). This goal will be accomplished via the following aims: (1) Identify novel features, multiclass (tissue) feature selection and segmentation of hard-to-detect tumors and associated abnormalities using multimodal MRIs from different imaging centers; (2) Enable robust segmentation of tumor, other abnormal and normal tissues and tacking of brain tumor by fusing atlas-based tumor segmentation (ABTS) and feature-based BTS (FBTS); (3) implement software integration into a widely available tool (3D Slicer) available via multiple NIH sponsored Resource Centers such as the Neuroimaging Analysis Center (NAC), the National Alliance for Medical Image Computing (NA-MIC), and the National Center for Image Guided Therapy (NCIGT), for wider dissemination of BTS tool; and (4) Validate and evaluate our integrated BTS tool to quantify improvements in the detectability, sensitivity and specificity, and corresponding errors.