基于CT和病理WSI的多尺度非小细胞肺癌术后复发风险智能量化研究

可切除非小细胞肺癌首选手术切除，但患者预后迥异，临床亟需精准量化复发风险指导治疗决策。课题组前期基于影像组学手工特征构建模型，可用于肿瘤诊疗评估，但精度待提高；近期课题组及其他团队发现，基于术后病理全切片数字化图像（WSI）和人工智能可表征肿瘤免疫微环境，为复发风险量化带来机遇。因此，本研究拟：（1）在宏观影像层面,提出图卷积神经网络（GCN）方法，表征肿瘤术前CT图像不同亚容积区域间的空间拓扑信息，量化肿瘤空间异质性；（2）在微观细胞层面，用GCN表征WSI上肿瘤细胞和浸润淋巴细胞间的空间拓扑关系，量化免疫微环境状态；（3）融合CT和WSI信息构建术后复发风险量化模型；（4）基于独立数据集验证和优化模型。本项目的实施，将构建多尺度复发风险智能量化模型，实现术后复发风险分层，为非小细胞肺癌术后诊疗方案制定提供量化工具，指导临床决策。

Surgical resection is generally the most preferable option for resectable non-small cell lung cancer (NSCLC), while the prognosis varies among different patients. Therefore, it is urgently needed to quantify the recurrence risk to aid clinical decision-making. Our prior studies demonstrated that hand-crafted feature-based radiomics prediction model could be used for tumor diagnosis and treatment response evaluation, whilst its prediction accuracy warranted to be improved. Recently, our group and other researchers found that combination of pathological whole slide image (WSI) and artificial intelligence has the potential to characterize the tumor immune microenvironment, providing an opportunity to quantify cancer relapse risk. Therefore, this study intends to: (1) at the macroscopic image level, propose a graph convolutional neural network (GCN) to characterize the spatial topological information between different sub-volume regions of preoperative tumor CT images to quantify the intratumoral spatial heterogeneity; (2) at the microscopic cell level, use GCN to characterize the spatial topological relationship between tumor cells and tumor infiltrating lymphocytes in postoperative tumor WSI to quantify the immune microenvironment; (3) combine CT and WSI information to construct a quantitative model of recurrence risk; (4) validate and optimize the prediction model in independent validation dataset. The implementation of this project will build a multi-scale intelligent quantification model of recurrence risk for risk stratification in NSCLC, and provide a quantitative tool in clinical decision-making for postoperative diagnosis and treatment planning.