Racially-associated MRI analysis and modeling for predicting aggressive prostate cancer

项目摘要 非洲裔美国人（AA）男性的前列腺癌（PCa）发病率和死亡率最高， 美国的前列腺多参数MRI（mpMRI）是一种非侵入性成像技术，可以灵敏地 通过整合解剖和功能信息来检测前列腺肿瘤。现行标准化方案 用于解释mpMRI的是前列腺成像报告和数据系统（PI-RADS）。然而，检测 癌性病变目前不能解释PI-RADS中种族相关的MRI特征。 我们的初步数据显示，AA和AA之间在检测具有临床意义的PCa（csPCa）方面存在显著差异。 当肿瘤位于移行区时，CA男性使用PI-RADS（分别为67%和80%，p=0.026）。 此外，AA和CA男性之间的PCa灌注（即Ktranss）存在明显差异， 通过定量动态对比增强MRI（qDCE）测量。当基于PI-RADS的解释是 结合为AA男性指定的Ktranss阈值， AA男性改善至76%，与CA男性无统计学差异（p=0.180）。 我们开发了一种名为P4的床旁便携式灌注体模，以提高qDCE的重现性 在不同的机构之间进行测量。基于P4的误差校正显著降低了 在两个研究所的三台MRI扫描仪上进行qDCE测量，并提高了Ktranss对csPCa的特异性 检测率从86%提高到93%。我们假设PCa诊断中的种族差异可以通过使用 基于P4的误差校正后的种族相关qDCE测量。 我们建议在加州大学洛杉矶分校的多机构环境中测试这一假设 （UCLA）和伯明翰的亚拉巴马大学（UAB）。我们的团队将收集并联系临床，放射学， 和组织病理学信息，使用患者特异性3D打印前列腺模具，软件配准， 根治性乳腺癌切除术前后的专家注释。高度策划的放射病理学数据集将是 用于（1）表征AA和CA组中与肿瘤微环境相关的qDCE测量， 在基于P4的误差校正后使用共定位定量放射学-病理学分析，（2）研究 种族相关的基于MRI的组织表征是否改善了侵袭性PCa的检测， 以及（3）开发种族/民族特定的深度学习模型，用于改进攻击性PCa的检测。 当目标成功实现时，AA和CA男性中PCa的改进检测是 与常规策略相比，预期减少检测侵袭性PCa的种族差异。

PROJECT SUMMARY African American (AA) men have the highest incidence and mortality rate from prostate cancer (PCa) in the United States. Prostate multi-parametric MRI (mpMRI) is a non-invasive imaging technique that can sensitively detect prostate tumors by integrating anatomical and functional information. The current standardized scheme for interpreting mpMRI is the Prostate Imaging Reporting and Data System (PI-RADS). However, detecting cancerous lesions currently does not account for racially associated MRI characteristics in PI-RADS. Our preliminary data showed a significant difference in detecting clinically significant PCa (csPCa) between AA and CA men using PI-RADS when the tumors are in the transition zone (67% vs. 80%, respectively, p=0.026). In addition, there was a distinctive difference in the PCa perfusion (that is, Ktrans) between AA and CA men, when measured by quantitative dynamic contrast-enhanced MRI (qDCE). When PI-RADS-based interpretation was combined with the Ktrans threshold value specified for AA men, the csPCa detection rate in the transition zone in AA men was improved to 76%, becoming not statistically different from that in CA men (p=0.180). We developed a point-of-care portable perfusion phantom named P4 to improve the reproducibility of qDCE measurement across different institutes. The P4-based error correction significantly reduced the variability in qDCE measurement across three MRI scanners in two institutes and improved the specificity of Ktrans for csPCa detection from 86% to 93%. We hypothesize that the racial disparity in PCa diagnosis can be reduced by using racially associated qDCE measurement after P4-based error correction. We propose to test this hypothesis in a multi-institutional setting at the University of California, Los Angeles (UCLA) and the University of Alabama at Birmingham (UAB). Our team will collect and link clinical, radiologic, and histopathologic information using patient-specific 3D-printed prostate molds, software registration, and expert annotation before and after radical prostatectomy. The highly curated radiology-pathology dataset will be used (1) to characterize the qDCE measurement associated with tumor microenvironment in AA and CA groups, using co-localized quantitative radiology-pathology analyses after P4-based error correction, (2) to investigate whether the racially associated MRI-based tissue characterization improves the detection of aggressive PCa, and (3) to develop the race/ethnicity-specific deep learning model for the improved detection of aggressive PCa. When the Aims are successfully accomplished, the improved detection of PCa in both AA and CA men is anticipated, compared to conventional strategies, reducing the racial disparity in detecting aggressive PCa.