(PQC4) Habitats in Prostate Cancer

描述(申请人提供)：这项提案将针对PQC-4：“可以开发什么体内成像方法来描绘肿瘤的‘细胞类型’，定义为组成肿瘤及其微环境的每种不同细胞类型的身份、数量和位置？解决这一问题的理想系统应具有以下特点：1)图像和数据应来自人类患者；2)图像和细胞类型之间的关系应具有临床相关性；3)从癌和非癌体积内获得的数据应大量且平衡；4)图像数据应具有高质量且理想的多参数；以及5)组织学与放射图像的配准必须可行。这些标准在接受主动监测(AS)监测的前列腺癌患者中得到满足。迈阿密大学(UM)有大量的AS患者，前列腺癌患者定期和常规进行多参数磁共振成像(MP-MRI)，包括扩散成像(DWI)、动态对比增强(DCE)和T2加权成像(T2W)序列作为标准护理(SOC)。这些图像被融合到经直肠超声(TRUS)引导仪器进行活检定位。目前工作的单一目标是开发预测模型，该模型基于深刻的图像分析(“放射组学”)，结合定量组织学和免疫组织化学，从空间共同配准的体积中定义这种相互关系；从而定义产生MR图像数据的“细胞类型”。莫菲特癌症中心的研究人员率先将放射组学和预测性(类似分类器)建模应用于癌症。因此，这项工作将继续进行 两个相互关联的目标。在目标1中，UM将通过MAST试验产生MR图像、组织学、基因表达和临床数据：MRI引导的活组织检查选择主动监视与治疗。在目标2中，将在莫菲特进行信息学数据分析、数据库和分类器建模。对核磁共振图像的分析将使用“放射组学”方法，即从图像识别的栖息地提取432个大小、形状和纹理特征。这些图像将与用定量病理学分析的注册组织学图像相匹配，其中从每个细胞中提取32个特征以形成相似形态类型的簇，以及已知和假定的进展标记的IHC。根据这些数量标记，将开发训练和测试集分类器模型，以将MR定义的栖息地与其潜在的细胞类型混合联系起来。由于这将是一个庞大而无价的数据库，我们明确的意图是通过材料转让协议与研究界共享完整的数据集，这将允许替代数据挖掘模式。

DESCRIPTION (provided by applicant): This proposal will address PQC-4: "What in vivo imaging methods can be developed to portray the "cytotype" of a tumor defined as the identity, quantity, and location of each of the different cell types that make up a tumor and its microenvironment? An ideal system to address this question will have the following characteristics: 1) images and data should be obtained from human patients; 2) the relationship between imaging and cytotypes should have clinical relevance; 3) there should be a large amount and a balance in data obtained from within cancerous and non-cancerous volumes; 4) the image data should be of high quality and ideally multiparametric; and 5) registration of histology to radiographic images must be feasible. Such criteria are met in prostate cancer patients who are being monitored by active surveillance (AS). The University of Miami (UM) has a large AS population, and patients with prostate cancer are regularly and routinely imaged with multiparametric MRI (MP- MRI) that includes diffusion (DWI), dynamic contrast enhancement (DCE) and T2 weighted (T2w) imaging sequences as standard of care (SOC). These images are fused to a transrectal ultrasound (TRUS) guidance instrument for biopsy localization. The singular goal of the current work is to develop predictive models that define this interrelationshi based on profound image analyses ("radiomics") in combination with quantitative histology and immunohistochemistry from spatially co-registered volumes; thus defining the "cytotypes" giving rise to MR image data. Researchers at the Moffitt Cancer Center have pioneered the application of radiomics and predictive (classifier like) modeling to cancer. Thus, this work will proceed with two interrelated aims. In Aim 1, MR images, histology, gene expression and clinical data will be generated at UM via the MAST Trial: MRI- Guided Biopsy Selection for Active Surveillance versus Treatment. In Aim 2, informatics data analysis, databasing and classifier modeling will be undertaken at Moffitt. Analysis of MR images will use a "radiomics" approach, wherein 432 size, shape and texture features are extracted from image-identified habitats. These will be matched up to registered histology images analyzed with quantitative pathology wherein 32 features are extracted from each cell to form clusters of similar morphotypes, as well as IHC for known and putative progression markers. From these quantitative markers, training and test set classifier models will be developed to relate the MR-defined habitats to their underlying mixtures of cytotypes. Because this will be a large and invaluable data base, it is our explicit intention to share the complete dataset, with the research community through material transfer agreements, which will allow alternative data mining schema.