文章摘要

文章信息/Info

Title:: Study on self-supervised emphysema grading algorithm based on goddard scoring method

作者:: 韩云龙1; 2; 王苹苹1; 2; 卢绪香3; 杨毅1; 2; 丁鹏4△; 魏本征1; 2△; （1.山东中医药大学青岛中医药科学院，青岛 266112；2.山东中医药大学医学人工智能研究中心，青岛 266112；3. 山东中医药大学附属医院，济南 250011；4. 山东中医药大学第二附属医院，济南 250001）

Author(s):: HAN Yunlong1; 2; WANG Pingping1; 2; LU Xuxiang3; YANG Yi1; 2; DING Peng4; WEI Benzheng1; 2; （1. Qingdao Academy of Traditional Chinese Medicine, Shandong University of Traditional Chinese Medicine, Qingdao 266112，China；2. Medical Artificial Intelligence Research Center, Shandong University of Traditional Chinese Medicine, Qingdao 266112；3. Shandong University of Traditional Chinese Medicine Affiliated Hospital, Jinan 250011,China；4. The Second Affiliated Hospital of Shandong University of Traditional Chinese Medicine, Jinan 250001）

Keywords:: Chronic obstructive pulmonary disease; Emphysema; CT imaging; Self-supervised learning; EMA; 3D convolution

摘要:: 针对肺气肿智能化诊断高度依赖高质量标注数据、图像空间信息复杂及特征提取不足等问题，本研究基于Goddard评分法设计了一种肺气肿分级算法。首先，利用SimSiam框架进行自监督学习，以解决对大量高质量标注数据的依赖；其次，引入连续3D卷积模块和高效多尺度注意模块（efficient multi-scale attention，EMA），通过整合上肺野、中肺野及下肺野的信息捕捉肺部图像中的关键空间信息，以提升模型在处理复杂肺部CT图像时的特征提取能力和识别精度。实验结果显示，在识别肺气肿存在、轻度肺气肿与无肺气肿、肺气肿严重程度的分级任务中，模型准确率分别为88.79%、83.44%、57.4%。结果表明，本算法在肺气肿识别和分类任务中表现良好，具有一定的临床意义。

Abstract:: Aiming at the intelligent diagnosis of emphysema highly depending on high-quality annotation data, complex image spatial information and insufficient feature extraction, we designed an emphysema classification algorithm based on Goddard scoring method.Firstly, the algorithm utilized the SimSiam framework for self-supervised learning to address the dependency on a large volume of high-quality annotated data. Then, the continuous 3D convolution module and the efficient multi-scale attention (EMA) module were introduced, to capture the key spatial information of lung images by integrating the information of upper, middle and lower lung lobes, to improve the feature extraction ability and recognition accuracy of the model were processing complex lung CT images.The experimental results showed that in the grading task of the emphysema presence, mild and no emphysema, and the severity of emphysema, the accuracy of the model was 88.79%, 83.44%, and 57.4%, respectively.The result indicates that this algorithm performs well in the emphysema recognition and classification, and has certain clinical significance.