Multiresolution-fractal modeling for brain tumor detection

描述(由申请人提供)：PI的长期研究目标是开发一种全功能的自动化、健壮的CAD工具，用于随着时间的推移准确地分割和跟踪儿童脑肿瘤体积。注意：目前在脑肿瘤体积分割方面的实践包括在多模式MRI中手动跟踪和分割可疑的肿瘤区域，这既耗时又费力，而且可能不精确。为了减少认知后遗症，当代的治疗方案采用了风险适应疗法，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是否存在转移疾病。因此，如果不提高对肿瘤体积和分类等因素的了解，儿童癌症治疗结果的进一步改善不太可能实现。此外，这种自动化的体积计算和跟踪工具将作为跟踪脑肿瘤患者的辅助标记具有价值。反过来，这将帮助医生做出关于手术计划、关键放射治疗计划修改、治疗领域修改、局部控制、转移疾病部位和治疗后反应评估的重要患者管理决策。然而，这种自动化和精确的肿瘤体积分割CAD工具的开发需要解决一些挑战，例如难以检测的脑肿瘤(手术后残留小、强化差、多个病灶和形状不规则)和异常(手术导致的水肿、坏死和较大的切除空洞)的检测和分类。该项目旨在开发、测试和评估创新的技术和工具，这些技术和工具将有助于脑肿瘤和一些特定异常的基于特征的检测、分割和分类。}该项目的具体目标是：1)样条多分辨率小波-分形特征提取；2)依赖于MR序列的特征融合和肿瘤/异常大小和体积的确定，以改进检测；3)优化特征融合，以改进肿瘤、组织和异常分类；以及4)算法测试和验证。{如果成功，我们的方法将允许以更高的精确度自动计算脑肿瘤和异常，这可以提供对疾病的快速、客观、可重复性和易于报告的评估。该项目的结果将对儿科神经放射学实践产生立竿见影的影响，为评估和解释脑肿瘤及相关异常提供准确、客观和一致的方法。 公共卫生相关性：该项目旨在开发、测试和评估新的基于特征的算法，用于健壮、准确和可重现的脑肿瘤和其他异常检测和分类。这样的识别和分类将被用来获得难以检测的脑肿瘤和异常的精确分割。我们将难以发现的脑肿瘤定义为小(手术后残留)、低强化、多个病灶和形状不规则的病变，以及因手术而导致的水肿、坏死和较大的切除空洞。该算法能够可靠、准确地计算分割的肿瘤体积，可作为脑肿瘤患者随访的辅助标记物。这种肿瘤体积量化方法也将直接应用于临床前手术计划和治疗试验，从而产生新的治疗策略和设备。该项目的结果将对神经放射学实践产生直接影响，为评估和解释脑肿瘤提供准确、客观和一致的方法。

DESCRIPTION (provided by applicant): The PI's long-term research goal is to develop a fully functional automated robust CAD tool for accurate pediatric brain tumor volume segmentation and tracking over time. Note the current practice in brain tumor volume segmentation involves manual tracing and segmentation of suspected tumor areas in multimodality MRI which is time consuming, labor intensive, and may be imprecise. In an effort to reduce cognitive sequelae, contemporary protocols employ risk-adapted therapy in which risk stratification is based on volume of residual tumor after surgical resection and presence of metastatic disease at diagnosis. Therefore, further improvement in cancer treatment outcome in children is unlikely to be achieved without improved knowledge of tumor volume and classification among other factors. In addition, such automated volume computation and tracking tool would be of value as an adjunct marker in following up patients with brain tumors. This will, in turn, help the physicians to make important patient management decisions about surgery planning, critical radiation treatment planning modifications, treatment field modifications, localized control, sites of metastatic disease and post therapy response evaluation. However, development of such automated and precise tumor volume segmentation CAD tool requires solution to a few challenges such as detection of hard-to-detect brain tumor (small residual after surgery, poorly enhanced, multi foci and irregularly shaped) and abnormalities (edema, necrosis, and larger resection cavity due to surgery) detection and classification. This project aims at development, testing, and evaluation of innovative techniques and tools that will assist feature-based detection, segmentation and classification of brain tumor and a few specific abnormalities.} The specific aims of this project are: 1) Spline-multiresolution wavelet-fractal feature extraction; 2) MR sequence-dependant feature fusion and tumor/abnormality size and volume determination for improved detection; 3) Optimized feature fusion for improved tumor, tissue and abnormality classification; and 4) Algorithm testing and validation. {If successful, our method will allow for the automatic computation of brain tumors and abnormalities with improved accuracy, which can provide a rapid, objective, reproducible, and easily reported assessment of the disease. The results obtained from this project will have immediate impact in pediatric neuroradiology practice by providing an accurate, objective, and consistent way to evaluate and interpret brain tumors and associated abnormalities. PUBLIC HEALTH RELEVANCE: This project aims at development, testing, and evaluation of novel feature-based algorithms for robust, accurate and reproducible brain tumor and other abnormalities detection and classification. Such identification and classification will then be used to obtain precise segmentation of hard-to-detect brain tumors and abnormalities. We define hard-to-detect brain tumor as lesions that are small (residual after surgery), poorly enhanced, multi foci and irregularly shaped and abnormalities as edema, necrosis, and larger resection cavity due to surgery respectively. The algorithms capable of reliably and accurately computing segmented tumor volume would be of value as an adjunct marker in following up patients with brain tumors. Such a tumor volume quantification method would also have direct application in pre-clinical surgery planning and therapy trials leading to novel treatment strategies and devices. The results obtained from this project will have immediate impact in neuroradiology practice by providing an accurate, objective, and consistent way to evaluate and interpret brain tumors.