结合解剖特征框架的婴幼儿脑MR图像分割与定量分析方法研究

Infant brain disease incidence is being with increasing tendency recent years, automatic/semi-automatic image segmentation and quantitive analysis, can effectively reduce missed-diagnosis and misdiagnosis of infant brain diseases. This research project aims to low quality complex infant MR brain images, the following three aspects will be discussed: 1) Research on image enhancement processing based on adaptive variable-order fractional differential method; 2) Face the different level problems of infant image segmentation, under the frame of utilizing medical anatomical characteristics, research on precise segmentation methods for brain tissue, deep brain structure and local lesion regions. The following methods will be included: convex functional MS model combined with local probability distribution, position associated Bayesian framework MRF model, combining gray and shape constraint non-rigid registration segmentation, and graph theory maximum flow/minimum cutting based on expert knowledge; 3) Research on quantitive calculation of infant cerebral cortex complexity base on fractal dimension and surface quantitative analysis; Infant brain structure morphology quantitative analysis based on shape feature description and manifold learning method; Construct brain statistical information based on limited sample of infant groups at different growth stage. The research project will apply the newest methods of information processing fields to medical image diagnosis, provide accurate comprehensive quantitative information for the diagnosis of infant brain diseases. The research also has certain scientific significance to improve the accuracy and scientificness for diagnosis of infant brain diseases, and the medical research of study of the growth rule of infant brain.

婴幼儿脑部疾病的发生率呈现上升趋势，自动/半自动的图像分割与定量分析，可以有效降低患儿脑病的漏诊和误诊。针对低质量复杂婴幼儿MR脑图像，研究基于自适应变阶次分数阶微分的图像增强方法；面向不同层次婴幼儿脑图像分割问题，研究脑组织、深层脑结构和局部病灶的精确分割方法，在结合脑部医学解剖特征框架下，研究结合局部概率分布的凸泛函MS模型、引入位置关联贝叶斯 MRF模型、结合灰度与形状约束的非刚性配准分割、融入专家知识的图论最大流/最小切等分割新方法；研究分形维和曲面分析的婴幼儿大脑皮层复杂度定量计算，基于形状特征描述、流形学习的婴幼儿脑结构形态定量分析，构建基于有限样本的幼儿群体不同生长阶段脑统计信息。课题将信息处理领域新方法充分应用到医学影像诊断中，为医生诊断婴幼儿脑部疾病提供准确全面的定量信息，对于提高患儿脑病诊断的准确性和科学性、探讨幼儿脑生长发育规律的医学研究都具有一定的科学意义。

婴幼儿时期是人类大脑发育的关键阶段，这一时期易患各种脑部疾病，近年来婴幼儿罹患各种脑病的几率呈逐年上升趋势。婴幼儿脑影像呈现对比度低、灰白质逆转等特点，使得婴幼儿脑图像自动/半自动的分割与定量分析更具挑战性。.本项目针对婴幼儿脑部MR图像成像质量低的特点、以及婴幼儿脑部疾病辅助诊断的各关键环节开展研究工作。在婴幼儿脑部MR图像增强处理方面，将分数阶微分方法应用于婴幼儿脑MR图像增强，提出基于自适应非整数步长分数阶微分、融合非局部均值信息、基于形态成分分析等改进分数阶微分方法，有效地改进了经计算后分数阶微分数值，更有效地增强婴幼儿脑MR图像。在面向不同层次的婴幼儿脑图像分割方面，提出MRF-MAP框架下融合概率图谱先验和非参数估计的分割、结合局部均值与全局均值的改进水平集模型的方法、基于核函数和局部空间信息的改进FCM算法、基于非对称多尺度图像块和稀疏表达的分割、基于改进CNN的分割算法等，有效提高了婴幼儿脑图像分割的准确性和稳定性。在婴幼儿脑形态分析与计算方面，提出基于球谐函数重建的婴幼儿大脑分形维数计算、基于光程函数方程的婴幼儿脑皮层厚度测量方法、基于改进张量法的主曲率估计及图切割方法的主曲率矫正、基于有限差分法主曲率估计的大脑皮层脑沟底线提取。研究结果与解剖学理论相符，可以反映大脑皮层表面复杂性的局部变化，有利于发现某些脑部疾病引起的大脑结构变化，为大脑皮层形态分析提供参数计算依据。.本项目构建了0~2岁婴幼儿脑MR影像数据库，在面向婴幼儿脑部计算机辅助诊断的关键环节提出一系列新方法，这些研究成果的取得，可为医生诊断婴幼儿脑部疾病提供准确全面的辅助信息，对于提高婴幼儿脑部疾病诊断的准确性、为脑发育研究提供有效的计算依据等方面都具有重要的科学价值与应用意义。