肝癌微波消融诊疗大数据分析及超声影像智能识别研究

It is of great significance to evaluate the curative effect, predict the survival and assist the clinical decision for malignant tumors using medical big data. During the microwave ablation (MWA) for hepatic cancers, because the data of diagnoses and treatments lack normalization and standardization, it is difficult to objectively and accurately assess the curative effects and predict the long-term survival. Moreover, ultrasound (US) image is one of the most important clinical image techniques in the diagnoses and treatments of liver cancers, however, it is difficult for US to achieve the precise recognition for liver malignancies and the residual tumors surrounding the ablated zone due to the absence of scanning standardization, low contrast resolution and high noise interference between the tumors and the adjacent tissue, and an absolute operator dependent technique. In order to solve these problems, this project will provide solutions basing on the methods of medical big data including standardized data collection, intelligent recognition and the techniques of radiomics. First, basing on definite data classification, collection and standardization, it is essential to develop and establish a clinical data base, which supports the entire, accurate and standard data, in order to establish a mathematical model for evaluation and prediction of the curative effects of liver cancers after MWA. Secondly, multi-layer US features of hepatic tumors are automatically extracted and analyzed by the convolutional neural network to perform the intelligent recognition of the liver cancers and the residual tumors surrounding the ablated zone. Finally, the mathematical models for evaluation and prediction of the curative effects and the intelligent recognition of the liver cancers and the residual tumors surrounding the ablated zone will be confirmed by a prospective clinical study, to improve the curative effects of percutaneous MWA for hepatic cancers and promote the medical big data and US radiomics in the clinical applications for the diagnoses and treatments of liver cancers.

利用医学大数据，对恶性肿瘤鉴别、疗效评估、生存预测及为临床提供辅助决策意义重大。肝癌微波消融诊疗中，数据缺乏规范标准，难以进行客观准确疗效评估预测。超声影像在肝癌诊疗中广泛应用，但影像采集标准未统一、肿瘤与周围组织图像对比分辨力低且噪声干扰大，且易受医生主观经验影响，精准识别肝癌边界和消融区周边残癌困难。本项目针对上述难题，拟基于医学大数据的标准化采集、智能化识别和影像组学技术提供解决方案。首先，明确数据分类、格式标准与采集规范，完成肝癌微波消融诊疗临床数据库，从完整、准确、规范的数据中提取并组合特征，构建肝癌消融疗效评估预测模型；其次，利用卷积神经网络深度挖掘并自动分析肝癌边界及消融区周边残癌的多层次超声影像特征，探讨其智能识别规律；最后，通过前瞻性设计的临床研究，验证疗效评估预测模型及超声影像智能识别规律，提高肝癌微波消融疗效并推动医学大数据和超声影像组学在临床肝癌诊疗中的应用。

肝脏良恶性肿瘤早期鉴别以及肝癌消融治疗后残癌或复发肿瘤及时诊断和处理对提高肝癌患者疗效和远期生存非常关键，医学大数据和人工智能方法有望对此提供重要辅助信息。本研究搭建了满足医学大数据要求的肝癌微波消融临床诊疗信息一体化数据库，采用深度残差网络建立回归模型，实现了适合临床使用的肝癌微波消融疗效评估及预后预测；利用深度卷积神经网络和全连接网络构成的双分支深层神经网络，分别提取肝癌及消融区周边残癌的超声影像特征，构建其智能识别模型，拓展了超声影像在肝癌诊断和消融疗效评估中的临床应用。