基于经颅超声多模态影像信息融合的帕金森病早期辅助诊断模型研究

项目名称： 基于经颅超声多模态影像信息融合的帕金森病早期辅助诊断模型研究

项目编号： No.61471231

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 无线电电子学、电信技术

项目作者： 施俊

作者单位： 上海大学

项目金额： 83万元

中文摘要： 帕金森病（PD）严重威胁中老年人群身体健康，其早期诊断至关重要。经颅超声（TCS）成像对PD早期诊断具有较高的敏感性和特异性，基于TCS的PD早期辅助诊断研究已是热点。由于单一模态TCS的辅助诊断性能尚不能满足临床要求，本项目提出融合TCS结构成像和多普勒功能成像的PD早期辅助诊断模型，研究多视图学习理论与方法，对多模态影像特征进行从结构到功能的数据融合与整合计算分析。然而，TCS不同影像特征之间时序不对应、特征维数差距大，需要针对多视图特征的不一致性问题研究新的多视图学习理论与方法。因此，本项目基于协同稀疏表示、多核学习、深度学习等理论和方法，研究结构化联合协同稀疏模型、稀疏多核核熵成分分析理论与方法，对TCS多模态特征进行降维与融合；研究概率分布空间的多层多核支持测度机的理论与方法，对TCS多模态影像特征进行模式分类；最终研究建立基于TCS多模态影像信息融合的PD早期辅助诊断模型。

中文关键词： 多模态融合；经颅超声；帕金森病；计算机辅助诊断；多视图学习

英文摘要： The Parkinson's disease (PD) has been a great threat to middle-aged and elderly population, and its early diagnosis is crucial for future treatment. Transcranial sonography (TCS) based computer-aided diagnosis (CAD) for early detection of PD has become a hot topic due to its high sensitivity and specificity for early PD diagnosis. However, the performance of single modality TCS based CAD for PD is still far from real clinical practice. In this project, we propose a multi-modality TCS based CAD, which fuses together the B-mode structure imaging and Doppler function imaging for early diagnosis of PD. The features extracted from structural and functional modalities will be integrated with multi-view learning methods. However, the mis-alignment in time between the two different modality features and the large discrepancys between their feature dimensionalities make it difficult for most of the existing method to be able to handle feature fusion of the two modalities. Therefore, we propose to study new multi-view learning paradigms for this specific problem. Based on the theories and methods of co-sparse representation, multiple kernel learning and deep learning, we study the structural joint co-sparsity model, and the theory and method of sparse multiple kernel entropy component analysis so as to reduce the dimensionalities of multi-modality features and fuse them together. We also study the theory and method for multi-layer multiple kernel support measure machine in the space of probability distributions to perform classification with multi-modality features. Finally, we study multi-modality TCS based CAD for early PD with these new multi-view learning methods.

英文关键词： Multi-modality Fusion;Transcranial Sonography;Parkinson's Disease;Computer-aided Diagnosis;Multi-view Learning