基于影像组学的乳腺癌超声诊疗关键问题研究

Breast cancer is one of malignant tumors with the highest incidence rate in women in the worldwide. Ultrasound technique has become a vital imaging method in screening and diagnosis for breast cancer. A new challenge is proposed by the current individual medical model towards the comprehensive diagnosis of the diseases by using ultrasonic imaging, genomes, biological information and clinical data. With the comprehensive analysis of the imaging and molecular and biological characteristics, applying the advanced radiomic analysis technique, the aim of this project is to explore a novel early diagnosis and prognosis evaluation method for breast cancer using multi-modal ultrasound imaging.. We will attempt to acquire multi-level and multi-parameter characteristics from multi-modal ultrasound imaging based on previous established ultrasound image database, which will be expanded and renewed. Then we will extract and select significant diagnostic features or the combination of these features from the histopathological and molecular biological levels. All of these characteristics and features will be analyzed quantitatively. Based on the above researches, the machine learning and data mining approach will be employed to comprehensively analyze the susceptible genes, key genetic information for invasion, as well as risk factors for breast cancer, to establish a model for the early quantitative diagnosis and risk evaluation, which will be validated, clinical evaluated and further optimized. Finally, we aim to establish a basic theoretical framework for the multi-modal breast cancer ultrasound radiomics study and to obtain an objective and quantitative model for the early diagnosis and prognosis evaluation. This study and this model will improve the early diagnostic rate and prediction accuracy for breast cancer.

乳腺癌是世界范围内的女性首发恶性肿瘤，乳腺超声检查是乳腺癌筛查诊断的重要影像学方法，当前个体化医疗模式对超声影像与基因组、生物信息、临床数据融合的综合诊断提出了新挑战。本项目的目标是借助宏观影像学与分子生物学特性的综合分析运用，应用先进的影像组学分析技术研究超声多模态早期诊断乳腺癌和预后评估的新方法。.本研究将基于前期工作所建立的并不断扩充更新的超声图像数据库，获取超声多种模态图像的多层次、多参数特征，从病理组织学和分子生物学水平出发，寻找并筛选出有意义的诊断特征或特征组合，进行量化分析。在此基础上利用机器学习和数据挖掘方法，融合乳腺癌的易感基因、侵袭性关键基因信息，以及乳腺癌风险因素，建立乳腺癌的早期量化诊断和风险评估模型，并对其进行验证、临床评估和进一步优化。最终建立超声多模态乳腺癌影像组学研究的基础理论框架，获得早期诊断和预后评估的客观量化模型，提高乳腺癌早期诊断率和预后预测精度。

本项目按照原计划顺利开展，原计划拟完成的五个研究内容均融入到每项研究方向中，具体如下：①乳腺多模态超声数据的标准化与完善：扩充了多模态乳腺超声影像数据库，并对其进行标准化和完善，同时建立全国多中心协作团队，制定了数据统一采集标准，进一步实现了万例以上级别的乳腺多模态超声数据库；②乳腺癌病理类型及分子分型相关超声影像组学研究：主要针对乳腺癌四种亚型、浸润性小叶癌和三阴性乳腺癌的多模态超声特征进行分析，提取筛选了多模态超声中的关键特征，用于模型建立，并对所建立的模型进行验证，展示出了多模态超声在乳腺癌病理和分子分型中的评估价值；③乳腺癌转移复发相关超声影像组学特征研究：主要包含两方面，一方面是关于评估乳腺癌腋窝淋巴结转移的研究，分析了腋窝淋巴结转移及非转移之间的常规超声特征的差异，在此基础上，应用人工智能深度学习技术研究设计了“DLRU模型”，逐步预测乳腺癌腋窝淋巴结转移风险，可使一部分低风险患者免于过度治疗，以上研究有助于乳腺癌的个体化腋窝管理；另一方面是关于乳腺癌复发相关超声特征的研究，该研究探索性的应用多模态超声评估乳腺癌患者的预后风险, 为乳腺癌预后提供了有力的参考依据；④在完成原计划的基础上，本项目还进行了针对乳腺癌新辅助化疗疗效相关的lncRNA进行筛选、研究北方汉族人群中VEGF单核苷酸多态性与乳腺癌的关系等小规模探索性研究。本项目资助下，共发表论文23篇，其中SCI收录16篇，累计影响因子80.728，单篇最高影响因子8.143；国家发明专利3项（1项已授权）；新获国自然面上项目1项，青年项目3项。