Potential of Radiomics and AI in the prediction of breast cancer risk and mutation status in high risk patients with confirmed mutation or calculated high risk status (PRo-mics-BrCa)

致病性基因突变被确定为显著增加患乳腺癌的终生风险，最显著的是BRCA 1和BRCA 2基因中的任一个。这些基因的有害突变会在受影响的家庭中产生遗传性乳腺癌-卵巢癌综合征。BRCA 1/2突变并不常见，而乳腺癌相对常见，因此这些突变仅占女性乳腺癌病例的5%至10%。随着全数字化乳腺成像（乳腺X线摄影和MRI）的引入，扩展的基于成像的表型和随后的多变量扩展表型-基因型相关性研究开始实现，以揭示乳腺癌风险的新的基于特定成像的指标。该应用旨在提供一套强大的工具，用于基于乳腺成像提取多尺度组织表征，特别关注时间变化和不对称分析，应用于过去十年在德国高风险人群中采集的大规模乳腺成像数据集的亚群，覆盖个体成像体积（> 6，000名女性，平均>4次检查）。所有这些妇女都接受了DNA检测，其中一部分因可疑发现而接受了额外的临床检测。我们将与德国遗传性乳腺癌和卵巢癌联合会合作访问并统计分析所有这些数据。由于DNA检测无法在更大的筛查人群中完全实现，并且已知的乳腺癌风险基因只能解释4分之1的已确定的遗传性乳腺癌风险病例，因此我们将测试来自动态对比增强乳腺MRI和全视野数字乳腺X射线摄影（FFDM）的许多定量成像生物标志物，并将这些成像生物标志物及其时间发展与乳腺癌发病率相关联，与致病基因突变的存在以及已确定的遗传性乳腺癌风险有关。这种相关性将产生一种基于成像的亚表型分类，对未来监测计划中的扩展分层具有潜在意义。成像生物标志物将包括形态学描述符、体积乳房组织组成和密度、组织异质性和不对称性、对比度增强模式以及其纵向变化等参数。我们将进行放射组学分析以及最先进的卷积神经网络深度学习方法提取乳腺组织组成和密度的形态学特征以及背景增强，以及与BRCA 1或2突变患者癌症发病率的关系，较不频繁的突变和计算的乳腺癌风险升高。总的来说，我们的目标是开发一种先进的工具集，以改善乳腺癌的早期检测，并在推导新的假设方面，扩大乳腺癌的风险。

Pathogenic gene mutations were identified that significantly increase the lifetime risk of developing breast cancer, most prominently in either of the genes BRCA1 and BRCA2. Harmful mutations in these genes produce a hereditary breast-ovarian cancer syndrome in affected families. Mutations in BRCA1/2 are uncommon, and breast cancer is relatively common, so these mutations account for only five to ten percent of all breast cancer cases in women. As full-digital breast imaging, both mammography and MRI, has been introduced, an extended imaging based phenotyping and subsequently multivariate extended phenotype-genotype correlation studies come into reach in order to reveal new specific imaging based indicators for breast cancer risk. This application aims at providing a powerful set of tools for extracting multiscale tissue characterization based on breast imaging with particular focus on temporal change and asymmetry analysis, applied to a subpopulation of a large-scale breast imaging dataset acquired over the last ten years in a high-risk population in Germany covering individual imaging volumes (>6,000 women with on average >4 examinations). All of these women were offered DNA testing and a part received additional clinical testing due to suspicious findings. We will access and statistically analyze all of this data in collaboration with the German Consortium for Hereditary Breast and Ovarian Cancer. Since DNA testing cannot be completely realized in a larger screening population, and the known breast cancer risk genes only explain 1 in 4 of the identified hereditary breast cancer risk cases, we will test a number of quantitative imaging biomarkers from dynamic contrast-enhanced breast MRI and from full-field digital mammography (FFDM), and correlate those imaging biomarkers and their temporal developments with breast cancer incidence, with the presence of pathogenic gene mutations as well as with the identified hereditary breast cancer risk. This correlation will produce an imaging-based subphenotype classification with potential implications for extended stratification in future surveillance programs. Imaging biomarkers will include, among other parameters, morphological descriptors, volumetric breast tissue composition and density, tissue heterogeneity and asymmetry, contrast enhancement patterns, as well as the longitudinal changes thereof. We will perform the Radiomics analysis as well as state-of-the-art convolution neural network deep learning approaches extraction for morphological features of breast tissue composition and density as well as background enhancement and the relationship to cancer incidence in patients with BRCA 1 or 2 mutation, less frequent mutations and calculated elevated risk for breast cancer. Overall, we aim at developing an advanced toolset for improved early detection of breast cancer and at deriving novel hypotheses with respect to extended breast cancer risk.