基于影像-病理组学对胰腺癌精准诊断及预后评估的研究

Providing individual diagnosis and treatment for pancreatic cancer (PC) has been the major requirement with the coming of the precision medicine. A number of bottlenecks of imaging diagnosis for PC come out. Firstly, small PC is difficult to be detected. Secondly, PC often accompanying chronic pancreatitis leads to the low accuracy of T staging. Thirdly, it is difficult to be staged the lymphatic metastasis and perineural invasion of PC. Lastly, predicting survival time and prognosis is almost impossible. Now it is urgent to find a diagnostic method for solving the bottlenecks and getting the precision diagnosis. In our preliminary study, we have explored some of the high correlation characteristics of PC between radiomics and pathomics about interstitial fibrosis. At the same time, we find that the model showed a good discrimination to PC and normal pancreas, yielding an accuracy of 98.97%. We also have find the same result about the discrimination to positive and negative lymph node, yielding an accuracy of 97.14%.. This project will investigate the utility a radio-path-omics analysis of thousands of features quantifying tumor radiologic and pathologic images. First, we will try to find a high correlation between features by Apriori correlation algorithm. Next, the prediction model will be established for lymphatic metastasis and perineural invasion. Then the logistic regression model will be established and generate the Nomogram figure based on the above results. Combined with the COX proportional hazards models, we will provide a radiologic decision reporting for PC lead to precision diagnosis.

为胰腺癌（pancreatic carcinoma，PC）患者提供个体化诊疗方案已成为精准医疗迫切需求。影像诊断PC主要瓶颈为对小胰癌诊断敏感性低，PC与胰腺炎的混杂导致T分期准确率不高，无法精准评估胰周淋巴结转移和神经浸润，无法预测患者生存和预后。我们前期工作已经探索出PC的影像组学与肿瘤间质纤维化病理组学的某些高关联特征，所建立的神经网络模型预测PC的准确率为98.97%，预测淋巴结转移的准确率为97.14%。. 本课题拟运用组学的方法，高通量提取影像-病理-组学特征，通过Apriori关联算法，进一步探索两者间高相关特征，建立肿瘤、淋巴分期和神经浸润预测模型；根据上述研究结果再建立PC诊断的Logistic回归模型并生成Nomogram图；结合COX比例风险模型对患者生存预测，为每一位患者制定影像决策化报告，实现精准诊断。

为胰腺癌（pancreatic carcinoma，PC）患者提供个体化诊疗方案已成为精准医疗迫切需求。影像诊断PC主要瓶颈为对小胰癌诊断敏感性低，PC与胰腺炎的混杂导致T分期准确率不高，无法精准评估胰周淋巴结转移和神经浸润，无法预测患者生存和预后。.基于此，本项目完成以下方面工作。（1）基于CT和MRI，对胰腺和肿瘤的全自动分割，实现了胰腺和肿瘤组织全自动分割模型。（2）基于正常胰腺组织和胰腺癌组织病理HE图像的精准分割，实现胰腺组织和胰腺癌组织各病理成分的精准分割。（3）基于CT影像组学对胰腺癌淋巴结转移无创预测，建立了胰腺癌淋巴结转移CT组学模型，实现了组学水平的精准预测。（4）基于CT图像全自动无创诊断胰腺癌淋巴结转移，建立了胰腺癌淋巴结转移的全自动CT诊断模型，突破了影像医生常规影像学方法无法诊断淋巴结转移的缺陷，克服了医生之间评价一致性差的难题，实现了术前CT图像全自动、无创、精准诊断，减轻了影像医生的阅片压力，为实现精准诊断，指导临床治疗和早癌筛查提供重要的决策信。（5）胰腺癌T分期影像和病理的相关性研究。该研究发现CT测得肿瘤直径更接近HE大切片组织的肿瘤直径，三者均显著小于肿瘤大体直径。HE切片镜下肿瘤直径较大体直径更加准确。并且我们获得CT与镜下直径的精准换算公式，实现术前无创精准判断T分期。