面向肺疾病检测的胸腔CT影像分割研究

发布时间：2018-06-20 12:36

本文选题：同态学处理 + 肺野分割　；参考：《电子科技大学》2013年博士论文

【摘要】：肺癌是当今世界人数死亡最多的癌症，在早期对它进行检测、诊断和治疗是提高肺癌患者生存率的重要手段。肺癌的早期征状是以小肺结节的形态出现，，故对肺结节的及早检出与及时治疗对挽救肺癌患者生命具有重要意义。随着医学CT(Computed Tomography))技术的进步，医学CT检查获得影像数据的大量增加，其能提供更多器官、组织信息的同时，也给医生带来了极大的读片工作负担。为提高医生的工作效率、减轻工作强度和克服读片中人为因素的影响以及提高对肺疾病的检出率，急需研究对胸腔CT影像中各种肺疾病进行自动检测的方法。而研究这样的方法，其首要的任务就是对胸腔CT影像中肺各组织进行正确的分割。由此，本文以胸腔CT影像为研究对象，以肺疾病自动检测为目的，通过结合人体的肺解剖结构、CT成像机理等知识，对肺的各组织、病变的分割方法展开研究。研究大致内容如下：从胸腔CT影像正确分割肺是进行肺疾病自动检测的重要步骤与首要任务，由此研究了一种结合同态学处理和CT阈值分割相结合的肺野分割算法。它在肺解剖知识模型的引导下，能够对有叶裂的肺野、分割粘连的肺野和有纵隔边缘凹陷的肺野均能进行正确处理和分割。针对有高密度近胸膜肺结节肺野，研究了一种基于先验形状约束的活动轮廓模型的肺野分割的方法。首先它对已分割的肺野形状进行分类，并对这些形状进行分类学习获得其PCA(Principal ComponentAnalysis)形状模式向量；然后通过该先验向量模式与活动轮廓相结合的模型迭代拟合来完成对肺野的分割。研究表明采用该方法对较规范的有边缘高密度病变的肺野分割是完全可行的。为更好地解决有高密度近胸膜肺结节肺野的分割问题，本文进一步研究利用相邻肺野的形状相似特征来正确分割该类肺野的方法。首先对胸腔CT影像中的肺野形状形成的相似流形和对肺野PCA流形进行了研究。然后研究通过肺野流形上点所表达的肺野关系采用流形插值重构肺野形状，再进行变形配准来减小误差的肺野分割方法。研究结果表明它是一种可行的分割方法，同时从分割结果的准确性和敏感性、特异性结果可看出它能分割除肺尖和肺底外有边缘病变肺野的正确区域。对平扫CT、低剂量CT存在着的大量噪声的影像,开展了通过组合滤波、医学影像增强、分割和分数阶微分增强能力等方面的研究，最后研究出了先用分数阶微分算子对肺影像增强,再用局部最优阈值进行血管分割的肺血管分割方法。研究表明该方法可有效地提取血管网络并得到丰富的血管细节，对比传统肺血管分割方法可知它有更为准确的肺血管分割能力。针对成像、重建噪声和部分容积效应的影响以及肺结节病变导致的胸腔CT影像中组织与肺结节病灶之间存在的边界模糊情况，研究了一种结合四邻域连接权的脉冲耦合神经网络结合先验形状能量函数的主动轮廓分割候选肺内结节和近胸膜肺结节的方法。研究表明它是一种切实可行且行之有效的候选肺结节分割方法。在本文的最后部分，阐述了用胸腔CT影像对肺疾病进行检测的发展趋势，以及下一步将要开展的研究工作。然后对论文研究工作的主要内容、创新点进行了总结并对未来的研究工作做了进一步的展望。
[Abstract]:Lung cancer is the most common cancer in the world today . It is an important means to detect , diagnose and treat lung cancer in the early stage .

A lung field segmentation algorithm combined with homomorphism and CT threshold segmentation is studied in this paper .

In this paper , a kind of lung field segmentation method based on a priori shape constraint is studied for lung field with high density subpleural lung nodules . Firstly , it classifies the shape of lung field which has been divided , and classifies these shapes to obtain Principal Component Analysis ( PCA ) shape pattern vector ;
The segmentation of lung field is completed by using the model iterative fitting of the prior vector mode and the active contour .

In order to solve the problem of lung field division of lung nodules with high density near pleural cavity , the authors further study the method of correctly dividing lung field by using the shape similarity feature of adjacent lung fields . First , the lung field shape formed by lung field shape in thoracic CT image is studied . The results show that it is a feasible method of segmentation , and the accuracy and sensitivity of the segmentation result are obtained .

In this paper , we study the image of large amount of noise in plain scan CT and low - dose CT . Finally , we study the method of pulmonary vascular segmentation by means of combination filtering , medical image enhancement , segmentation and fractional order differential enhancement . The results show that the method can effectively extract the vascular network and obtain abundant vascular details , and it can be seen that it has more accurate pulmonary vascular segmentation ability compared with traditional pulmonary vascular segmentation method .

Based on the effect of imaging , reconstruction of noise and partial volume effect and the blurring of the boundary between tissue and lung nodule lesions in thoracic CT images caused by pulmonary nodule lesions , a method of segmenting candidate lung nodules and proximal pleural nodules by a pulse - coupled neural network combining four - neighborhood connection weights is studied . The results show that it is a feasible and effective method for the segmentation of lung nodules .

In the last part of this paper , the developing trend of detecting lung diseases with chest CT image and the research work to be carried out next step are described . Then the main contents and innovation points of the research work are summarized and the future research work is looked forward .
【学位授予单位】：电子科技大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：TP391.41;R816.41

【参考文献】