三维血管图像技术的探索与研究

The diseases that relate to blood vessels such as cardiovascular diseases have become one of the greatest threats of human. Nowadays as the development of computer aided diagnosis technology, three dimensional imaging techniques of blood vessels have been playing a more and more important role in diagnosis and treatment of vessel diseases. Because some problems such as the poor performance of feature enhancement and the low accuracy and computation efficiency of vessel segmentation exist in the previous methods, the computer aided diagnosis is not well performed. This project proposes an accurate and efficient three-dimensional vessel image processing technique, exploring the feature enhancement method and vessel segmentation approach, to increase the accuracy of computer aided diagnosis. Three research points are included here: (1) Feature enhancement technique based on vesselness and structure decomposition; (2) Vessel segmentation based on space reflection; (3) Feature selection method based on Gaussian Mixture Model. These three points aim to solve the three problems: junction suppression, difficulty in training 3D supervised segmentation model and the poor ability to optimize feature space. The three points are correlated and enhance for each other, realizing the accurate detection of 3D vessel structure. This project is supposed to build an accurate and efficient 3D vessel image processing system, which provides important evidence for vessel disease diagnosis, promoting the development of precision medicine and automated diagnosis.

与血管相关的疾病如心脑血管疾病已经成为人类健康的最大威胁之一。随着计算机辅助诊断技术的发展，三维血管的图像分析技术在血管疾病的诊治中扮演着越来越重要的角色。由于现有的方法中存在特征加强效果不佳、血管分割准确性及计算效率低等问题，导致辅助诊断存在偏差。本项目拟提出一种准确而高效的三维血管图像处理技术，针对特征加强方法与血管分割技术进行系统的探究，以提高辅助诊断准确性。本项目主要包括以下三个研究点：（1）基于血管度及结构分解的特征加强技术；（2）基于空间映射的分割方法；（3）基于高斯混合模型的特征选择方法。这三个研究点分别对应解决以下三个问题：交叉结构抑制、三维监督式分割模型训练难度大以及特征空间优化能力不足。三个研究点层层深入、互相加强，共同实现血管结构在三维图像中的精确检测。本项目拟输出一个准确而高效的三维血管图像处理系统，为血管疾病的诊治提供重要依据，促进精准医疗、自动诊疗技术的发展。

血管等复杂结构的检测是医学图像处理领域的挑战。目前的血管检测技术存在着对细微结构的检测不够准确、运算较为繁琐、计算资源消耗大等问题。.针对当前血管结构检测甚至医学图像处理存在的问题，本项目开展了血管分割、特征选择等算法的研究，并取得如下进展：1) 为了提升血管检测的准确率，本项目提出了基于卷积神经网络的血管分割方法； 2) 为了加强对细小结构及断裂结构的检测，本项目提出了形变U形网络 (Deformable U-Net)提取血管结构；3) 为了精简特征、提高算法准确度，本项目提出了MinE-RFE特征选择方法，去除冗余信息以及提高效率；4) 为了平衡准确率与特征数目，实现精准特征定位，本项目提出了WERFE等特征选择算法；5) 本项目的成果不仅仅局限于血管的检测，还可以推广到其他结构的检测和疾病的诊断，为图像技术在医学的应用提供了重要的技术前提和保障。.本项目如期并按计划完成任务。项目负责人以第一或者通讯作者在Knowledge-Based Systems、Briefings in Bioinformatics、Neurocomputing等期刊发表论文多篇，培养博士、硕士研究生6名。

内容获取失败，请点击重试