基于混合模型的多模态医学图像特征点非刚性配准算法研究

多模态医学图像可提供丰富的医学信息，能帮助医生制定更加合理的诊疗方案。多模态医学图像特征点非刚性配准旨在匹配图像中具有相同诊疗意义的点或区域，具有配准精度高、计算速度快的优点，是医学图像处理领域的研究热点。但是，现有算法存在以下问题：不能准确寻找退化点集之间的对应关系，无法有效配准退化点集；不能有效描述退化点集的局部空间结构，无法精确配准复杂分布的退化点集。围绕上述问题，本项目的主要研究内容是：基于t分布混合模型，研究针对退化点集的配准算法，提高其鲁棒性和准确性，并获得配准参数的闭合解，论证其收敛性，以形成完整的算法模型。在此基础上，针对复杂分布的退化点集，进一步研究基于Dirichlet过程混合模型的点集局部空间结构描述算法，并用于重构配准参数，提高退化点集的配准准确性和精度。拟研究的特征点配准算法将提高特征点非刚性配准的准确性、精度和鲁棒性，具有重要的学术研究价值和实际应用价值。

Multimodal medical images can help doctors develop more rational treatment programs owing to their rich information. Feature point set registration for multimodal medical images, which aims to match the corresponding points or regions between the multimodal medical images, is a key component in medical analysis because of its high accuracy and fast implementation. However, the classical point set registration algorithms have the following shortcomings: poor robustness to degradation such as noise, outliers and missing data and disability to find the correspondences and translation between the degenerated point sets; disability to describe the spatial structure of complex point sets for non-rigid registration. To address the above shortcomings, we will focus on non-rigid point set registration for multimodal medical images, including the studies of: a non-rigid point set registration model for degenerated point sets using the Student’s-t mixture model, and the spatial structure describer of complex point sets using the Dirichlet process mixture model. The non-rigid point set registration algorithm will improve the precision and robustness of multimodal medical image registration and has great theoretical and practical value.