Radiomics and Pathomics to predict upstaging of DCIS

摘要 乳腺导管原位癌是一组异质性的肿瘤性病变，通常是 通过筛查乳房X光检查发现。检查通常包括经皮(核心)活检(Bx) 组织学证实，然后是多参数磁共振成像(MpMRI)，然后是保乳切除，以及 辅助放射治疗。大约20%-25%的核心Bx确诊的DCIS患者被降级为侵袭性 切除组织的病理基础上的癌。如果事先知道这一点，就会采取更激进的外科手术 干预，包括前哨淋巴结活检进行腋窝分期。此外，另有20%-25%的患者被 有低风险疾病和目前的想法是，这样的妇女可能会有更好的结果在积极 监测环境，这一点正在临床试验中进行测试。它的最终目标和总体影响 项目是使用机器学习来识别生化(SA1)或成像(SA2)生物标记物以及它们的 联合(SA3)区分惰性和侵袭性DCIS，通过切除后的抢占舞台来确定 活组织检查。 这项工作要检验的主要假设是低氧和低氧相关蛋白的表达 (HRP)可以区分好斗的DCIS和更懒惰的DCIS，这可以用于决策支持。 通过免疫组织化学(IHC)，hrp的表达是最佳的特征，我们已经部署了方法 用于多路传输的IHC，以及使用机器学习进行高级分析的方法(病理组学)。我们有 研究还表明，可以使用机器学习从mpMRI中识别乳腺癌内的缺氧栖息地。 (放射组学)。因此，我们建议使用核心活检的病理组学和mpMRI的放射组学来确定 手术前DCIS中低氧生境的存在和程度预测手术后的后续进展 切除手术。这项工作将在病理组学目标1和放射组学目标2中进行，目标3将开发 放射-病理组学联合预测指标。每个目标将包括：(A)用于训练、调整和 测试；(B)预期的内部和外部队列，以进行严格的验证。对于回溯性研究， 我们已经确定了604例DCIS女性患者接受了核心Bx、mpMRI和手术病理检查 在过去的10年里在莫菲特。内部前瞻性研究将每月收集约6名已同意的女性 全面的癌症护理®方案，并在莫菲特完成了他们的完整检查。外部验证队列将 在加州大学旧金山分校和Advent Health计入。 在这项工作结束时，我们将为DCIS开发一个风险模型，该模型可以在手术前部署 从一端的主动监测到更广泛的外科干预，引导决策 在另一端。这有望为后续的介入试验奠定基础。此外，还包括 低氧作为一个中心假说，具有很高的潜力来阐明这一自然历史的组成部分 疾病。

Abstract Ductal carcinomas in situ (DCIS) of the breast are a heterogeneous group of neoplastic lesions that are usually detected by screening mammography. Workup generally includes a percutaneous (core) Biopsy (Bx) for histologic confirmation, followed by multiparametric MRI (mpMRI), followed by breast-conserving excision, and adjuvant radiation. Approximately 20-25% of patients with core Bx-confirmed DCIS are upstaged to invasive carcinoma upon pathology of resected tissue. Foreknowledge of this would dictate a more aggressive surgical intervention, including sentinel node biopsy for axillary staging. Further, another 20-25% of patients are judged to have low-risk disease and current thought is that such women may have better outcomes in an active surveillance setting, and this is being tested in clinical trials. The ultimate goal and the overall impact of this project is to use machine learning to identify biochemical (SA1) or imaging (SA2) biomarkers, as well as their combination (SA3) to discriminate indolent from aggressive DCIS, as determined by upstaging upon excisional biopsy. The major hypothesis to be tested in this work is that hypoxia and expression of hypoxia-related proteins (HRPs) can discriminate aggressive from more indolent DCIS, and that this can be used for decision support. Expression of HRPs is optimally characterized by immunohistochemistry (IHC), and we have deployed methods for multiplexed IHC, as well as methods for advanced analytics using machine learning (pathomics). We have also shown that hypoxic habitats within breast cancers can be identified from mpMRI using machine learning (radiomics). We thus propose to use pathomics of core biopsies and radiomics of mpMRI to determine the presence and extent of hypoxic habitats in DCIS prior to surgery to predict subsequent upstaging after surgical resection. This work will be performed in Aim 1 for pathomics and Aim 2 for radiomics, and Aim 3 will develop combined radio-pathomics predictors. Each aim will contain: (a) retrospective arms for training, tuning, and testing; and (b) prospective internal and external cohorts for rigorous validation. For the retrospective studies, we have identified 604 cases wherein women with DCIS obtained core Bx, mpMRI, and surgery with pathology at Moffitt in the last 10 years. Internal prospective studies will accrue ~6 women/month who have consented to the total Cancer Care® protocol and who have their complete workup at Moffitt. External validation cohorts will be accrued at UCSF and at Advent Health. At the end of this work we will have developed a risk model for DCIS that can be deployed prior to surgery to guide decisions along the spectrum from active surveillance at one end to more extensive surgical intervention at the other. This is expected to lay a foundation for subsequent interventional trials. Additionally, the inclusion of hypoxia as a central hypothesis has high potential to illuminate components of the natural history of this disease.