基于深度学习的继发性肺结核CT辅助诊断模型构建及验证

doi:10.19982/j.issn.1000-6621.20230356

摘要/Abstract

摘要：

目的: 评价基于深度学习的继发性肺结核CT辅助诊断模型在临床应用中的价值。方法: 回顾性收集2018年12月至2023年4月在重庆市公共卫生医疗救治中心接受胸部CT平扫的2004例患者的病例资料,分为肺部正常组(544例)、普通肺部感染组(526组)和继发性肺结核组(934例)。按照随机分组(通过R语言的sample函数实现训练集和测试集的完全随机分组)的方式,将数据集划分为训练集(1402例,70.0%)和测试集(602例,30.0%)。所有图像采用肺野自动分割算法,获得肺野区域。进一步采用BasicNet和DenseNet算法进行三组间的分类研究。采用曲线下面积(area under curve,AUC)、敏感度、特异度和准确率评价模型的分类性能。最后,在测试数据中,将最优模型与3位不同年资的放射科医生的诊断结果进行比较。结果: 602例独立测试集中,DenseNet模型的性能优于BasicNet模型,两种模型的平均AUC、敏感度、特异度和准确率分别为92.1%和89.4%、79.7%和74.0%、89.4%和86.6%、86.2%和83.3%。其中,DenseNet模型的诊断性能优于低年资医生(准确率分别为90.7%和89.1%,Kappa=0.677),与中年资和高年资医生的诊断水平(准确率分别为90.7%、92.2%和95.3%,Kappa值分别为0.746和0.819)保持高度一致性。结论: DenseNet 模型能较准确地识别继发性肺结核,与放射科中年资医师的诊断水准相当,可以作为继发性肺结核的辅助诊断工具。

关键词: 结核, 体层摄影术, X线计算机, 人工智能, 模型, 统计学

Abstract:

Objective: To develop a deep learning-based auxiliary diagnostic model for secondary tuberculosis using CT scans and evaluate its clinical applicability. Methods: A retrospective collection was conducted on clinical data of 2004 patients who underwent chest CT scans at the Chongqing Public Health Medical Center from December 2018 to April 2023. The patients were divided into three groups: secondary tuberculosis (934 patients), ordinary lung infection (526 patients), and normal lungs (544 patients). Using a completely random sampling method, the dataset was divided into a training set (1402 patients, 70.0%) and a test set (602 patients, 30.0%). An automatic lung field segmentation algorithm was applied to isolate the lung field in all images. BasicNet and DenseNet classification algorithms were used for categorize the three groups. The discriminative performance of the model was evaluated using metrics such as area under curve (AUC), sensitivity, specificity, and accuracy. Finally, the optimal model was compared with three radiologists of different years of experience using testing data. Results: Using 602 samples in an independent test set, the DenseNet model demonstrated superior performance compared to the BasicNet model. They achieved an average AUC, sensitivity, specificity, and accuracy of 92.1% vs. 89.4%, 79.7% vs. 74.0%, 89.4% vs. 86.6%, and 86.2% vs. 83.3%, respectively. The diagnostic performance of the DenseNet model was superior to that of young doctors (accuracy: 90.7% and 89.1%, Kappa=0.677) and exhibited high diagnostic consistency with middle and highly experienced radiologists without any significant difference (accuracy: 90.7%, 92.2% and 95.3%, Kappa=0.746, 0.819). Conclusion: The DenseNet model can accurately identify secondary tuberculosis, achieving a competency level similar to a middle experienced radiologist, making it a potential auxiliary diagnostic tool for secondary tuberculosis.

Key words: Tuberculosis, Tomography, X-ray computed, Artificial intelligence, Models, statistical

中图分类号:

R521
R445

刘雪艳, 王芳, 李春华, 唐光孝, 郑娇凤, 王惠秋, 李玉蕊, 王佳男, 舒伟强, 吕圣秀. 基于深度学习的继发性肺结核CT辅助诊断模型构建及验证[J]. 中国防痨杂志, 2024, 46(3): 279-287. doi: 10.19982/j.issn.1000-6621.20230356

Liu Xueyan, Wang Fang, Li Chunhua, Tang Guangxiao, Zheng Jiaofeng, Wang Huiqiu, Li Yurui, Wang Jia’nan, Shu Weiqiang, Lyu Shengxiu. Construction and evaluation of a CT-based deep learning model for the auxiliary diagnosis of secondary tuberculosis[J]. Chinese Journal of Antituberculosis, 2024, 46(3): 279-287. doi: 10.19982/j.issn.1000-6621.20230356

图/表 12

图1

图2

图3

图4

图5

图6

图7

表1

图8

图9

图10

图11

参考文献 26

[1]	舒薇, 刘宇红. 世界卫生组织《2023年全球结核病报告》解读. 结核与肺部疾病杂志, 2024, 5(1):15-19. doi:10.19983/j.issn.2096-8493.2024006.
[2]	高雯琬, 郭建琼, 李同心, 等. 《结核病免疫治疗专家共识(2022年版)》要点解读. 结核与肺部疾病杂志, 2023, 4(3): 194-197. doi:10.19983/j.issn.2096-8493.20230035.
[3]	伍小英, 谭守勇, 罗少霞, 等. 广州市2009—2013年肺结核流行状况分析. 广东医学, 2014, 35(21): 3400-3405.
[4]	Brady AP. Error and discrepancy in radiology: inevitable or avoidable? Insights Imaging, 2017, 8(1):171-182. doi:10.1007/s13244-016-0534-1.
[5]	Zhou Q, Zhu W, Li F, et al. Transfer Learning of the ResNet-18 and DenseNet-121 Model Used to Diagnose Intracranial Hemorrhage in CT Scanning. Curr Pharm Des, 2022, 28(4): 287-295. doi:10.2174/1381612827666211213143357.
[6]	马利利, 饶会林, 唐飞, 等. 肺隐球菌病患者肺部CT影像学表现及临床病理特征分析. 临床军医杂志, 2022, 50(12): 1310-1312, 1316. doi:10.16680/j.1671-3826.2022.12.30.
[7]	von Schacky CE, Sohn JH, Liu F, et al. Development and Validation of a Multitask Deep Learning Model for Severity Grading of Hip Osteoarthritis Features on Radiographs. Radiology, 2020, 295(1): 136-145. doi:10.1148/radiol.2020190925. pmid: 32013791
[8]	Shen L, Wang L, Liu C, et al. Community-acquired pneumonia: Trends in and research on drug resistance and advances in new antibiotics. Biosci Trends, 2021, 15(5): 266-275. doi:10.5582/bst.2021.01342. pmid: 34483225
[9]	Yang W, Ma Y, Wang J, et al. Fungal pneumonia manifesting as cavitary lesions in a critically ill elderly patient. J Infect Dev Ctries, 2019, 13(12):1170-1173. doi:10.3855/jidc.11265.
[10]	重庆市医学会放射医学分会胸心学组, 重庆市医学影像医疗质量控制中心. 新型冠状病毒肺部感染放射学检查、诊断规范重庆专家共识. 陆军军医大学学报, 2023, 45(7): 593-605. doi:10.16016/j.2097-0927.202301092.
[11]	中华人民共和国国家卫生和计划生育委员会. WS 288—2017肺结核诊断. 2017-11-09.
[12]	Shan F, Gao Y, Wang J, et al. Abnormal lung quantification in chest CT images of COVID-19 patients with deep learning and its application to severity prediction. Med Phys, 2021, 48(4):1633-1645. doi:10.1002/mp.14609.
[13]	Cheng Z, Qin L, Cao Q, et al. Quantitative computed tomography of the coronavirus disease 2019 (COVID-19) pneumonia. Radiol Infect Dis, 2020, 7(2):55-61. doi:10.1016/j.jrid.2020.04.004.
[14]	Ouyang X, Huo J, Xia L, et al. Dual-Sampling Attention Network for Diagnosis of COVID-19 From Community Acquired Pneumonia. IEEE Trans Med Imaging, 2020, 39(8): 2595-2605. doi:10.1109/TMI.2020.2995508.
[15]	Zhao X, Qi S, Zhang B, et al. Deep CNN models for pulmonary nodule classification: Model modification, model integration, and transfer learning. J Xray Sci Technol, 2019, 27(4): 615-629. doi:10.3233/XST-180490.
[16]	Skrede OJ, De Raedt S, Kleppe A, et al. Deep learning for prediction of colorectal cancer outcome: a discovery and validation study. Lancet, 2020, 395(10221): 350-360. doi:10.1016/S0140-6736(19)32998-8.
[17]	Chen PT, Wu T, Wang P, et al. Pancreatic Cancer Detection on CT Scans with Deep Learning: A Nationwide Population-based Study. Radiology, 2023, 306(1):172-182. doi:10.1148/radiol.220152. pmid: 36098642
[18]	吴键, 侯代伦. 深度学习在肺结核影像诊断中的应用. 中国防痨杂志, 2022, 44(1): 91-94. doi:10.19982/j.issn.1000-6621.20210537.
[19]	Cao XF, Li Y, Xin HN, et al. Application of artificial intelligence in digital chest radiography reading for pulmonary tuberculosis screening. Chronic Dis Transl Med, 2021, 7(1): 35-40. doi:10.1016/j.cdtm.2021.02.001.
[20]	Kim TK, Yi PH, Hager GD, et al. Refining dataset curation methods for deep learning-based automated tuberculosis screening. J Thorac Dis, 2020, 12(9): 5078-5085. doi:10.21037/jtd.2019.08.34. pmid: 33145084
[21]	Qin ZZ, Ahmed S, Sarker MS, et al. Tuberculosis detection from chest x-rays for triaging in a high tuberculosis-burden setting: an evaluation of five artificial intelligence algorithms. Lancet Digit Health, 2021, 3(9): e543-e554. doi:10.1016/S2589-7500(21)00116-3.
[22]	Natarajan S, Sampath P, Arunachalam R, et al. Early diagnosis and meta-agnostic model visualization of tuberculosis based on radiography images. Sci Rep, 2023, 13(1): 22803. doi:10.1038/s41598-023-49195-x.
[23]	Ma L, Wang Y, Guo L, et al. Developing and verifying automatic detection of active pulmonary tuberculosis from multi-slice spiral CT images based on deep learning. J Xray Sci Technol, 2020, 28(5): 939-951. doi:10.3233/XST-200662.
[24]	马依迪丽·尼加提, 米日古丽·达毛拉, 张斌, 等. 基于CT深度学习的活动性/非活动性肺结核分类模型构建及验证. 暨南大学学报(自然科学与医学版), 2023, 44(1): 69-77, 86. doi:10.11778/j.jdxb.20220268.
[25]	Wang L, Ding W, Mo Y, et al. Distinguishing nontuberculous mycobacteria from Mycobacterium tuberculosis lung disease from CT images using a deep learning framework. Eur J Nucl Med Mol Imaging, 2021, 48(13): 4293-4306. doi:10.1007/s00259-021-05432-x.
[26]	吴树才, 王新举, 纪俊雨, 等. 基于深度学习卷积神经网络的肺结核CT诊断模型效能初探. 中华结核和呼吸杂志, 2021, 44(5): 450-455. doi:10.3760/cma.j.cn112147-20210108-00026.

模型/类别	曲线下面积	敏感度	特异度	准确率	精确率	F₁值
BasicNet
正常肺部	94.1	80.0	91.7	88.7	77.0	78.5
普通肺部感染	86.7	64.3	89.1	82.7	66.9	65.6
继发性肺结核	87.4	77.8	79.0	78.4	77.8	77.8
宏平均	89.4	74.0	86.6	83.3	73.9	74.0
微平均	88.9	74.9	84.8	82.2	74.8	74.9
DenseNet
正常肺部	96.7	92.9	90.8	91.4	77.8	84.7
普通肺部感染	89.0	68.8	90.2	84.7	70.7	69.7
继发性肺结核	90.6	77.5	87.1	82.4	85.0	81.1
宏平均	92.1	79.7	89.4	86.2	77.8	78.5
微平均	91.8	79.2	88.8	85.3	79.5	79.1