【作者】 李楠；

【导师】 刘继欣；

【作者基本信息】 西安电子科技大学 ， 生物医学工程， 2023， 硕士

【摘要】 哮喘是最常见的慢性疾病之一,在儿童时期的发病率和患病率较高。儿童哮喘治疗延迟或治疗过度均可能导致哮喘长期预后不良,严重影响患者的生活质量。然而,由于难以获得可逆性气流受限的客观数据,以及儿童哮喘严重程度的分级具有滞后性,儿童哮喘的早期诊断和严重程度评估一直是一个巨大的挑战。基于计算机断层扫描(computed tomography,CT)图像的肺气道分析为以气道结构异常为特征的肺部疾病的研究提供重要信息,特别是对于哮喘研究。现阶段的气道分割方法主要包括了纯手动标记、基于阈值或区域生长的方法、以及基于深度学习的方法。当前,基于深度学习的气道分割方法越来越受到研究者的关注。近年,深度学习中UNet与Transformer的组合模式不断刷新着细分任务的上限。UNet结合局部信息的低级特征图和表示全局信息的高级特征图,Transformer依靠注意力机制来模拟全局依赖关系,两者优势互补、相辅相成。然而,上述两种方法组合尚未应用到儿童气道分割中。针对上述问题及儿童气道分割的迫切需求,本研究整理归纳了常用的五种UNet与Transformer的组合模式,并从网络训练与气道分割两个层面对这五种组合模式进行了评估。实验结果发现UNet与Transformer的组合模式仍可适用于儿童气道分割,并且基于多尺度的学习模式更加适合儿童气道的分割,该种组合模式不仅训练耗时短,并且网络稳定性好,分割结果也表现优异。哮喘是一种伴有气体交换障碍的长期疾病,多数研究将其归结为气道阻塞。尽管有研究表明哮喘患者大气道结构的改变可能代表哮喘的早期阶段,但很少有研究关注患者大气道形态结构与大气道内空气流动不良的情况。在本研究中,我们提出了一种基于形态学的气道特征提取方法,此方法主要包括了气道分割、气道中心线获取、气道横截面的截取,以及形态特征的提取。我们使用该方法,提取了包括主气道、左主支气道和右主支气道在内的0至1代大气道的形态特征,结合定量CT测量在轻度哮喘患儿组、中度哮喘患儿组和重度哮喘患儿组间进行组间对比,并构建了哮喘药物治疗疗效的预测模型。组间对比结果发现,随着哮喘严重程度增加,患者的气管更加椭圆,提示患者可能存在气道较窄现象,并且气道壁厚不均匀,暗示着炎症的发生。此外,药物治疗疗效预测模型的结果也佐证了组间对比结果,表明哮喘严重程度与气道形态学存在潜在关系。上述研究结果表明气道形态测量对于完善3岁以下儿童哮喘的及时识别和早期治疗干预的潜在价值,为儿童哮喘早期诊断和严重程度评估的研究提供新的思路。更多还原

【Abstract】 Asthma is one of the most common chronic diseases,with a higher incidence and prevalence in childhood.Delayed or overtreatment of asthma in children may lead to poor long-term prognosis of asthma and seriously affect the quality of life of patients.However,the early diagnosis and severity assessment of childhood asthma has been a great challenge due to the difficulty in obtaining objective data on reversible airflow limitation and the lag in grading severity of childhood asthma.The analysis of pulmonary airway based on computed tomography(CT)images provides important information for the study of pulmonary diseases characterized by abnormal airway structure,especially for the study of asthma.At present,airway segmentation methods mainly include pure manual labeling,threshold or region-growing based methods,and methods based on deep learning.At present,airway segmentation based on deep learning has attracted more and more attention from researchers.In recent years,the combination model of UNet and Transformer in deep learning is constantly updating the upper limit of segmentation tasks.UNet combines the low-level feature diagram of local information with the high-level feature diagram of global information.Transformer relies on the attention mechanism to simulate the global dependency relationship.The two advantages complement each other.However,the combination of these two methods has not yet been applied to children’s airway segmentation.In view of the above problems and the urgent needs of children’s airway segmentation,this study summarizes five commonly used combination modes of UNet and Transformer,and evaluates these five combination modes from the two aspects of network training and airway segmentation.The experimental results show that the combination mode of UNet and Transformer can still be applied to children’s airway segmentation,and the multi-scale learning mode is more suitable for children’s airway segmentation.This combination mode not only takes less time in training,but also has good network stability and excellent segmentation results.Asthma is a chronic disease with impaired gas exchange,which most studies attribute to airway obstruction.Although some studies have shown that changes in airway structure in patients with asthma may represent the early stage of asthma,few studies have focused on the morphological structure of the airway and the poor air flow in the airway.In this study,we proposed an airway feature extraction method based on morphology,which mainly includes airway segmentation,airway centerline acquisition,airway cross section interception,and extraction of morphological features.We used this method to extract the morphological characteristics of atmospheric passages from 0 to 1 generation,including the main airway,the left main airway and the right main airway.Combined with quantitative CT measurement,we made comparison between the groups of children with mild asthma,moderate asthma and severe asthma,and built a prediction model of the therapeutic effect of asthma drugs.The results of comparison between the groups showed that as asthma severity increased,the trachea of patients became more elliptical,suggesting that patients may have narrower airways and uneven airway wall thickness,suggesting inflammation.In addition,the results of the drug therapy efficacy prediction model also supported the results of the comparison between the groups,suggesting a potential relationship between asthma severity and airway morphology.These results indicate the potential value of airway morphometry in improving the timely identification and early treatment and intervention of asthma in children under 3 years of age,and provide new ideas for the study of early diagnosis and severity assessment of childhood asthma.更多还原