光声-超声多模态影像组学模型智能预测类风湿性关节炎复发的方法研究

类风湿性关节炎（rheumatoid arthritis RA）为近年来发病率增高且严重影响患者生存质量的自身免疫疾病，规范治疗、预防复发对改善预后至关重要，RA治疗缓解后复发率高达50%-75%，而现有临床及影像学检查无法精确预测RA复发风险，因此亟需探索新型精准评估方法。申请人团队自主研发的新型光声/超声多模态成像设备可获取RA形态、功能等多模态图像信息，通过研究迁移学习和图卷积神经网络的深度学习方法分析多模态图像数据，有望实现多模态图像特异性特征的深度挖掘；通过分析多模态图像特征之间关联性的流形图结构、各模态图像特征与复发风险的相关性，建立RA复发智能预测模型。本项目拟应用新型光声/超声多模态成像评估RA，研究基于深度学习的影像组学关键技术以提取多模态影像组学特征，实现RA复发风险的精准预测评估，指导并优化临床诊疗路径，为多种临床表征复杂、影像精准诊断困难的疑难病种提供借鉴解决方案。

Recent years have witnessed a rapid increasement of the incidence of rheumatoid arthritis (RA), a common autoimmune disease, which had exerted a profound influence on the individual patients and social economics. Therefore, it is essential to implement standard treatment and prevent recurrence. The recurrence rate of RA is reported to be 50-75%, and there is no access to accurate prediction of recurrent risk using current clinical and imaging methods. Thus, new techniques for precise evaluation of RA recurrence risk are in urgent demand. The multimodality PA/US imaging system developed by our team can acquire functional and anatomical imaging of the joints of RA patients. Through deep learning method of transfer learning and graph convolutional neural network, the multimodal image data can be analyzed, facilitating the deep-mining of multimodal image features. By analyzing the manifold structures of the correlations of the multimodal image features, the relevance of the imaging characteristics and the recurrence risk, an intellegent predictive model of the RA recurrence risk can be established. In this study, the joints of RA patients will be evaluated by multi-modal PA/US imaging system, the high-throughput imaging features will be extracted through the key techniques of radiomics, to realize the precise prediction of RA recurrence risk. The aim of this study is to guide and optimize clinical pathways for the management of RA, provide potential solutions for certain types of diseases with complicated clinical phenotype leading to difficulties in accurate imaging diagnosis.