Evaluation of novel radiomics and machine learning-derived CT-biomarkers in patient stratification and therapy response assessment in pancreatic ductal adenocarcinoma

胰腺导管腺癌（PDAC）是最致命的癌症之一，其5年生存率仅为9%。此外，PDAC的发病率正在上升，预计到2030年，PDAC将在美国癌症相关死亡中占据第二位。PDAC的一个特殊标志是高度的基因组、转录组和代谢异质性，严峻的预后和特征性的肿瘤内和肿瘤间异质性强调了更精确的患者分层和治疗反应评估以提高PDAC生存率的必要性。然而，常规横断面成像方法在PDAC治疗反应评估和患者分层方面的价值有限。基于人工智能的算法可以帮助克服这个问题，并为个性化精确肿瘤学铺平道路。然而，以前在PDAC领域的研究最重要的是缺乏足够的队列规模，标准化和组织病理学特征的可用性。在此，我们的目标是为原发性可切除PDAC患者开发一个全面的分层模型，在迄今为止发表的最大的多机构和多国患者队列中结合最先进的放射组学和高端深度学习程序。此外，我们的目标是揭示新的综合成像衍生生物标志物的潜力，包括放射基因组学特征图谱，用于预测前瞻性COMPASS试验中纳入的晚期PDAC患者的缓解评估。该试验包括> 330例晚期PDAC患者。所有患者均接受了广泛的组织病理学和基因组检查（例如全基因组测序和RNA测序）。我们将最先进的机器学习和放射组学方法与突变和转录特征相匹配，以便为高级PDAC开发新的成像衍生生物标志物和放射基因组学特征图谱。通过我们提出的研究项目，我们的目标是开发先进的非侵入性成像衍生的生物标志物，以分别在原发性可切除和晚期PDAC患者中进行更好的患者分层和治疗反应评估。

Pancreatic ductal adenocarcinoma (PDAC) ranges among the deadliest cancers, reflected by a devastating 5-year survival rate of only 9%. Moreover, the incidence of PDAC is rising and it is projected to take over second place among cancer-related deaths in the United States by the year 2030. A particular hallmark of PDAC is the high degree of genomic, transcriptomic and metabolic heterogeneity.The grim prognosis and the characteristic intra- and intertumoral heterogeneity underline the necessity for a more precise patient stratification and therapy response assessment to improve survival in PDAC. However, routine cross-sectional imaging methods are of limited value with regard to therapy response assessment and patient stratification in PDAC. Artificial intelligence based algorithms could help to overcome this issue and pave the path towards individualized precision oncology. However, previous studies on this field in PDAC most importantly lack of sufficient cohort sizes, standardization and availability of histopathologic features. Herein, we aim to develop a comprehensive stratification model for patients with primary resectable PDAC, combining state of the art radiomics and high-end deep learning procedures in the so far published largest multi-institutional and multi-national patient cohort. Moreover, we aim to unravel the potential of novel comprehensive imaging-derived biomarkers, including radiogenomic feature maps, for prediction of response assessment in patients with advanced PDAC included in the prospective COMPASS trial. This trial includes > 330 patients with advanced PDAC. All patients received extensive histopathological and genomic work-up (e.g. whole genomic sequencing and RNA sequencing). We will match state-of-the art machine learning and radiomics approaches with mutational and transcriptional features in order to develop novel imaging-derived biomarkers and radiogenomic feature maps for advanced PDAC. With our proposed research projects, we aim to develop advanced non-invasive imaging-derived biomarkers for a better patient stratification and therapy response assessment in patients with primary resectable and advanced PDAC, respectively.