基于稀疏表示和压缩感知的多模态医学影像融合与超分辨率重建新方法

The modern medical imaging technology plays a most impotant role in the disease research and clinical diagnosis. However, it still has three bottlenecks, including the rather low space resolution, insufficient multimodal image fusion, and rather heavy calculating complexity for analysis and reconstruction. The above mentioned bottlenecks decrease the possibility to diagnose and treat the disease early. In order to solve above problems, this project is going to complete following studies: (1) the super resolution reconstruction for medical images based on the sparse representation (SR) and compressive sensing (CS) theory and technology; (2) new multimodal medical image and image sequence fusion methods both in 2 dimention and 3 dimention based on the SR based dictionary training; (3) the superpixel construction and segmentation for medical image based on the novel non Euclidean metric-geodesic distance and fuzzy distance; and (4) the applications on the above results to solve the difficulties in the diagnosis based on the medical imaging technology. The study and complement of this project may break through the above mentioned three bottlenecks and provide a useful technical support for the clinical diagnosis and treatment based on the medical imaging technology. This project is also very important to construct the theoretical system on the RS and CS based medical image processing methods.

现代医学影像技术在临床诊断与研究中具有极其重要的作用。但是，其尚存在空间分辨率不够高、多模态影像融合不够好和影像分析重建速度不够快等局限性，在一定程度上影响临床疾病的早期诊断和治疗。针对上述问题，本项目依据稀疏表示基本理论，以压缩感知、超分辨率重建和超像素技术为有力工具，拟系统研究：(1)基于稀疏表示和压缩感知的医学影像超分辨率重建方法；(2)基于稀疏表示字典训练和超像素技术的多模态医学影像及序列的二维与三维融合方法；(3)基于测地距离和模糊距离等非欧氏距离度量的医学影像超像素构造方法；(4)应用上述研究成果解决医学影像临床诊断的技术难题。本项目的完成，可以在不改变现行医学影像设备硬件条件的前提下，突破其尚存的空间分辨率、多模态影像融合和计算速度三个技术瓶颈，为临床诊断治疗提供技术支撑，且对于构建基于稀疏表示和压缩感知的医学影像处理新方法的理论体系具有重要意义。

现代医学影像技术在临床诊断与研究中具有极其重要的作用。但是，其尚存在空间分辨率不够高和多模态影像融合不够好的局限性，在一定程度上影响临床疾病的早期诊断和治疗。针对上述问题，本项目依据稀疏表示基本理论，深入研究了基于稀疏表示和压缩感知的医学图像超分辨率重建算法和多模态医学影像融合方法。具体成果包括：研究发展了一种基于字典学习和压缩感知的超分辨率重建新算法；提出了基于K-SVD字典训练算法和稀疏表示的多模态医学影像融合方法、基于ODL字典训练算法和稀疏表示的多模态医学影像融合方法及基于双稀疏字典的医学图像融合算法；将上述研究成果应用于医学影像临床诊断；针对图像处理中的噪声干扰与模糊问题，深入研究了图像预处理问题，有效解决了图像去噪声和去模糊问题。本项目的完成，对于在不改变现行医学影像设备硬件的前提下突破多模态影像融合技术瓶颈具有重要的应用价值，且对于构建基于稀疏表示和压缩感知的医学影像处理新方法的理论体系具有重要意义。