多模态腹部医学图像非刚性配准算法研究

Multi-modal abdominal imaging is an important method for the diagnosis and assessment of abdominal lesions. Image registration is an important technique to assist doctors in analyzing multi-modal images. Due to the structure differences of multiple organ and non-rigid large deformation in the abdomen, the commonly used multi-modal image registration algorithms do not meet the accuracy requirements for diagnosis. This project analyzes the similarity measures of classical multi-modal registration and use the model average theory to design a universal similarity measure by combining the advantages of various similarity measures. This universal similarity measure is developed for improving the registration accuracy for the multi-modal multi-organ of the abdominal images. Based on the analysis of the motion models of various organs in the abdomen, we study the transformation model for local non-rigid large deformation, and rebuild the global non-rigid transformation model for improving the registration accuracy based on the local transformation models. We also study the typical image segmentation algorithms. Then based on the abdominal image atlas, we design a multi-modal image joint segmentation-registration algorithm to further improve the registration accuracy. The accurate multi-modal non-rigid abdominal image registration is of great significance in reducing the average reading time of radiologists and improving the diagnostic rate of abdominal lesions.

多模态腹部成像是腹部病变诊断和评估的重要手段，图像配准是辅助医生分析多模态图像的重要技术。由于腹部存在多器官结构差异和非刚性大型形变等问题，常用的多模态图像配准算法已无法满足精度需求。本项目拟通过对经典的多模态配准相似性测度的分析，融合各类相似性测度优点，利用模型平均的思想设计一种针对腹部多模态多器官配准的普适性相似性测度，提升配准精度；通过对腹部各器官运动模型的分析，研究局部非刚性大型形变的空间变换模型，并基于各局部变换模型，重建用于提高配准精度的全局非刚性变换模型；研究成熟的图像分割算法，基于腹部图像图谱（atlas），设计和优化多模态图像联合分割配准算法，进一步提高配准精度。本项目所要解决的高精度腹部多模态图像非刚性配准问题，对降低放射科医生平均阅片时间并提升腹部病变的诊断率有重大意义。

多模态腹部成像是腹部病变诊断和评估的重要手段，图像配准是辅助医生分析多模态图像的重要技术。本项目针对腹部存在多器官结构差异和非刚性大型形变等问题，开展了多模态配准相似性测度、大形变器官或组织的空间变换模型、多模态图像联合分割配准等内容的研究，提出了基于结构特征描述子的匹配-配准联合优化方法、基于区域注意力机制的多模态图像分割-配准等方法。提出方法在腹部多模态图像配准上进行了验证，在大型形变区域得到了准确的配准结果，利用基于相位的结构特征描述子，解决了局部灰度差异等多模态图像配准中常见问题。此外，本项目研究的方法还直接应用在了新冠CT病灶定量分析中，为新冠病人的快速筛查和病程跟踪提供了技术支撑，使得该项目的研究形成了从理论到实践的闭环。