基于NAM和生物视知觉的医学图像分割算法研究

Medical image segmentation is not only the basis for high-level understanding of medical image analysis, but also it is a bottleneck of the current clinical applications. In view of the complexity of medical images, with the help and reference of biological visual perception mechanism, a new way of thinking to develop and improve the capability of the machine information processing and the cognitive computing is provided in this research. Based on the idea of the Non-symmetry and Anti-packing pattern representation Model (NAM) proposed by our research team and the theory of visual perception and by introducing the NAM theory to medical image segmentation, a new segmentation model for medical images which will be based on the NAM will be established. After representing the medical image by using our proposed NAM representation method, according to the different types of subpatterns and image patterns, then we will propose a novel NAM-based medical image segmentation method by designing some high-performance NAM-based splitting and merging rules and their corresponding data structures. Finally, an experimental system of medical image segmentation based on the NAM will be constructed. Theoretically, introducing the NAM theory to medical image segmentation is a new exploration with the help of visual perception mechanism. Clinically, it has significant practices in the medical field, such as accurate quantification diagnosis, formulation of surgical planning, visualization, pathological transformation tracking, evaluation of treatment effect, and so on.

医学图像分割不仅是高层的医学图像分析理解的基础，而且也是当前临床医学应用的瓶颈。借助和参考生物视知觉机理为发展和提高机器信息处理与认知计算能力提供了一种新的思路。针对医学图像的复杂性，基于本研究小组提出的非对称逆布局的模式表示模型(NAM)的思想，借助于视知觉理论，通过将NAM理论引入医学图像的分割中，拟首先建立基于NAM的医学图像分割模型。然后以医学图像的NAM表示方法为基础，针对不同的子模式类型和不同的图像模式，通过设计高效的基于NAM的分裂规则和合并规则以及与其相应的数据结构，拟提出一种新的基于NAM的医学图像分割方法。最后拟构建一个基于NAM的医学图像分割的实验系统。理论上借助于视知觉机理将NAM理论引入到医学图像的分割中是一个新的探索，临床上在医疗的精确量化诊断、手术计划的制定、可视化、病理变换的跟踪和治疗效果的评价等各个方面还具有重要的实际应用意义。

医学图像分割不仅是高层的医学图像分析理解的基础，而且也是当前临床医学应用的瓶颈。借助和参考生物视知觉机理为发展和提高机器信息处理与认知计算能力提供了一种新的思路。针对医学图像的复杂性，借助于“大范围首先”（Visual System Sensitive to Global Topological Properties，VSSGTP）的不变性知觉理论，通过将非对称逆布局模型（Non-symmetry and Anti-packing pattern representation Model，NAM）理论引入到医学图像的分割中，建立了一种新的医学图像分割模型，简称为NAM_VSSGTP模型，并给出了该模型的抽象化算法。以矩形子模式为例，提出了一种基于NAM_VSSGTP的医学图像分割算法。该算法的时间复杂度为O(Nα(N))，其中N为灰度图像用矩形NAM表示后的同类块总数，α(N)是Ackerman函数的反函数。实验结果表明：与流行的基于二元树(Binary Partition Tree，BPT)和基于四元树阴影编码（Quadtree and Shading-based Coding，QSC）的分割算法相比，在保持图像质量的前提下，基于NAM_VSSGTP的分割算法不仅具有更高的压缩比和更少的块数，而且能够显著提高医学图像的分割速度，因而是一种更好的图像分割方法。