肝癌的血管侵犯是一个逐步进展的过程，肝癌若伴大血管侵犯，预后极差；但是，若仅伴有微血管的侵犯则有可能获得根治性治疗的机会。如果能在微血管侵犯阶段进行积极的干预措施，则能明显改善肝癌患者预后。本课题采用多模态MR成像技术（包括多b值扩散成像、峰度成像、钆塞酸二钠增强前后T1 mapping成像）联合“相似度”树状图（包含传统的机器学习方法）及神经网络深度学习方法（我们提出一个基于掩码区域卷积神经网络的模型，以MRI多模态图像为输入，肝癌病灶自动检测，分割以及预测微血管侵犯这三个任务学习集成于一个网络，完成多任务学习的目的），并结合临床资料、肿瘤标记物及循环肿瘤细胞(CTCs)等手段，对肝细胞癌微血管侵犯进行回顾性和前瞻性研究，进一步与术后病理、全基因测序进行对照，建立术前预测肝癌微血管侵犯的智能诊断方法，为肝癌患者的个体化治疗及精准治疗提供依据。

Vascular invasion of HCC is a multi-step biological process. Patients with macrovascular invasive HCC inevitably have bad long-term survival outcomes. But if patients have only microvascular invasive HCC, they may get the chance to undergo curative treatments. So if we can predict microvascular invasion of patients with HCC, we can improve long-term survival outcomes of patients. In this study, we focus on multimodal MRI (including Diffusion weighted imaging with multiple b values, Diffusion kurtosis imaging and T1 mapping on gadoxetic acid-enhanced) combining tree diagram of similarity (including traditional machine learning method), neural network deep learning method (We proposed a Mask R-CNN based neural network model which takes inputs of multimodal magnetic resonance images. Different from prior networks in this field, our model integrates automatic HCC lesion detection, segmentation and MVI classification into one network which facilitates multitask learning), clinic factors, tumor markers and CTCs for the preoperative prediction of MVI, as well as to provide a clinically useful nomogram as a more direct and convenient tool to treat patients with individual and precise treatment and assist therapy decision-making in HCC.