
浏览全部资源
扫码关注微信
重庆市公共卫生医疗救治中心医学影像科,重庆
吕圣秀,E-mail: r202206@126.com
收稿日期:2025-04-29,
修回日期:2025-06-06,
纸质出版日期:2025-07-30
移动端阅览
李春华,刘雪艳,郑娇凤,等.基于CT影像的3D ResNet18模型在肺结核耐药性识别中的预测性能研究[J].陆军军医大学学报,2025,47(14):1676-1684.DOI:10.16016/j.2097-0927.202504117.
LI C H,LIU X Y,ZHENG J F,et al.Predictive performance of CT images-based 3D ResNet18 model for identifying tuberculosis drug resistance[J].J Army Med Univ,2025,47(14):1676-1684.DOI:10.16016/j.2097-0927.202504117.
李春华,刘雪艳,郑娇凤,等.基于CT影像的3D ResNet18模型在肺结核耐药性识别中的预测性能研究[J].陆军军医大学学报,2025,47(14):1676-1684.DOI:10.16016/j.2097-0927.202504117. DOI:
LI C H,LIU X Y,ZHENG J F,et al.Predictive performance of CT images-based 3D ResNet18 model for identifying tuberculosis drug resistance[J].J Army Med Univ,2025,47(14):1676-1684.DOI:10.16016/j.2097-0927.202504117. DOI:
目的
2
通过构建并验证一种基于胸部CT影像的深度学习模型,以正确识别耐药肺结核和药物敏感肺结核。
方法
2
回顾性纳入重庆市公共卫生医疗救治中心2019年1月至2022年12月诊治的722例继发性肺结核CT影像,根据药物敏感性试验结果,将其分为药物敏感肺结核组和耐药肺结核组,其中药物敏感肺结核357例,耐药肺结核365例。采用U-Net分割模型预先对CT影像的肺实质区域进行分割,然后按8∶2的比例随机划分为训练集和测试集。评估6种三维深度学习网络(3D Swin Transformer、3D ShuffleNet v2、3D ViT、3D MobileNet v2、3D DenseNet及3D ResNet18)对肺结核耐药性的识别性能,在训练集上通过五折交叉验证调整超参数,以构建最优深度学习模型。采用曲线下面积(area under the curve,AUC)、灵敏度、特异度、阳性预测值(positive predictive value,PPV)、阴性预测值(negative predictive value,NPV)以及F1值评价模型性能。由6名放射科医师对测试集CT影像进行耐药性识别,并与性能最优的深度学习模型进行性能比较。
结果
2
3D Swin Transformer、3D ShuffleNet v2、3D ViT、3D MobileNet v2、3D DenseNet及3D ResNet18预测耐药肺结核的AUC分别为0.583、0.704、0.698、0.758、0.736及0.841,3D ResNet18为性能最优深度学习模型,其灵敏度为0.935(95%
CI
:
0.880~0.987),特异度为0.642(95%
CI
: 0.492~0.757),阳性预测值为0.750(95%
CI
: 0.663~0.835),阴性预测值为0.896(95%
CI
: 0.809~0.976),AUC值为0.841,F1值为0.832。放射科医师F1值分别为0.571、0.450、0.675、0.623、0.617、0.635,3D ResNet18模型F1值均高于放射科医师。识别耐药肺结核结果最优的放射科医师的灵敏度、特异度、阳性预测值、阴性预测值分别为0.701(95%
CI
:0.605~0.802)、0.567(95%
CI
: 0.447~0.684)、0.651(95%
CI
: 0.549~0.757)、0.623(95%
CI
: 0.500~0.754),其灵敏度、阳性预测值和阴性预测值均低于3D ResNet18模型(
P
<
0.05)。类激活映射图表明3D ResNet18能够关注到关键病变区域。
结论
2
3D ResNet18 模型在肺结核耐药性识别中预测性能最优,有望辅助临床诊断耐药肺结核。
Objective To develop and validate a deep learning model based on chest CT images to accurately distinguish between drug-resistant (DR-TB) and -sensitive tuberculosis (DS-TB). Methods A retrospective study was conducted on 722 cases of confirmed secondary tuberculosis admitted in our center from January 2019 to December 2022. According to the results of antimicrobial susceptibility test
they were divided into 357 DS-TB cases and 365 DR-TB cases. Pre-existing U-Net segmentation model was employed to segment the lung parenchyma regions in CT images. The dataset was randomly partitioned into a training set and a testing set in an 8:2 ratio. Six 3D deep learning architectures (3D Swin Transformer
3D ShuffleNet v2
3D ViT
3D MobileNet v2
3D DenseNet
and 3D ResNet18) were employed to evaluate the discriminative efficiency between DS-TB and DR-TB. Hyperparameters were optimized by five-fold cross-validation on the training set to construct the optimal model. The performance of the constructed model was assessed using area under the curve (AUC)
sensitivity
specificity
positive predictive value (PPV)
negative predictive value (NPV)
and F1-score. Six radiologists independently evaluated DR-TB identification on the test set
and their performance was compared with the best-performing deep learning model. Resul
ts The AUC value in DR-TB prediction was 0.583
0.704
0.698
0.758
0.736
and 0.841
respectively
for 3D Swin Transformer
3D ShuffleNet v2
3D ViT
3D MobileNet v2
3D DenseNet
and 3D ResNet18. The 3D ResNet18 model demonstrated optimal performance
achieving a sensitivity of 0.935 (95%
CI
: 0.880
~
0.987)
a specificity of 0.642 (95%
CI
: 0.492
~
0.757)
a PPV of 0.750 (95%
CI
: 0.663
~
0.835)
an NPV of 0.896 (95%
CI
: 0.809
~
0.976)
an AUC value of 0.841
and a F1-score of 0.832. The radiologists got a F1-score of 0.571
0.450
0.675
0.623
0.617 and 0.635
respectively
and the F1-score of the 3D ResNet18 model is all higher than that of the radiologists. The highest-performing radiologist achieved sensitivity
specificity
PPV and NPV of 0.701 (95%
CI
: 0.605
~
0.802)
0.567 (95%
CI
: 0.447
~
0.684)
0.651 (95%
CI
: 0.549
~
0.757)
and 0.623 (95%
CI
: 0.500
~
0.754)
with all these values lower than those of the 3D ResNet18 model (
P
<
0.05). Class activation mapping showed that the 3D ResNet18 model could focus on key lesion areas. The class activation mapping demonstrated that the 3D ResNet18 model could effectively focus on critical lesion regions. Conclusion Our 3D ResNet18 model shows the best predictive performance in identifying DR-TB
and is expected to assist clinical diagnosis for DR-TB.
World Health Organization . Global tuberculosis report 2024 [EB/OL]. [ 2024-10-30 ]. https://publichealthupdate.com/global-tuberculosis-report-2024/ https://publichealthupdate.com/global-tuberculosis-report-2024/ .
KARKI M , KANTIPUDI K , YANG F , et al . Generalization challenges in drug-resistant tuberculosis detection from chest X-rays [J]. Diagnostics (Basel) , 2022 , 12 ( 1 ): 188 . DOI: 10.3390/diagnostics12010188 http://dx.doi.org/10.3390/diagnostics12010188 .
ZOU L P , KANG W L , GUO C H , et al . Treatment outcomes and associated influencing factors among patients with rifampicin-resistant tuberculosis: a multicenter, retrospective, cohort study in China [J]. Infect Drug Resist , 2024 , 17 : 3737 - 3749 . DOI: 10.2147/IDR.S467971 http://dx.doi.org/10.2147/IDR.S467971 .
CHEN M L , DODDI A , ROYER J , et al . Beyond multidrug resistance: Leveraging rare variants with machine and statistical learning models in Mycobacterium tuberculosis resistance prediction [J]. EBioMedicine , 2019 , 43 : 356 - 369 . DOI: 10.1016/j.ebiom.2019.04.016 http://dx.doi.org/10.1016/j.ebiom.2019.04.016 .
马璐瑶 , 郑秋婷 , 陆普选 , 等 . 基于文献分析的耐多药肺结核影像学研究进展 [J]. 新发传染病电子杂志 , 2021 , 6 ( 1 ): 65 - 72 . DOI: 10.19871/j.cnki.xfcrbzz.2021.01.015 http://dx.doi.org/10.19871/j.cnki.xfcrbzz.2021.01.015 .
MA L Y , ZHENG Q T , LU P X , et al . Pulmonary multi-drug resistant tuberculosis: a radiological review [J]. Electron J Emerg Infect Dis , 2021 , 6 ( 1 ): 65 - 72 . DOI: 10.19871/j.cnki.xfcrbzz.2021.01.015 http://dx.doi.org/10.19871/j.cnki.xfcrbzz.2021.01.015 .
FENG B , CHEN X M , CHEN Y H , et al . Solitary solid pulmonary nodules: a CT-based deep learning nomogram helps differentiate tuberculosis granulomas from lung adenocarcinomas [J]. Eur Radiol , 2020 , 30 ( 12 ): 6497 - 6507 . DOI: 10.1007/s00330-020-07024-z http://dx.doi.org/10.1007/s00330-020-07024-z .
崔效楠 , 叶兆祥 . 基于术前CT影像的肺结节病理类型预测模型研究 [J]. 中华放射学杂志 , 2024 , 58 ( 2 ): 243 - 244 . DOI: 10.3760/cma.j.cn112149-20231023-00309 http://dx.doi.org/10.3760/cma.j.cn112149-20231023-00309 .
CUI X N , YE Z X . Study on prediction model of pathological types of pulmonary nodules based on preoperative CT images [J]. Chin J Radiol , 2024 , 58 ( 2 ): 243 - 244 . DOI: 10.3760/cma.j.cn112149-20231023-00309 http://dx.doi.org/10.3760/cma.j.cn112149-20231023-00309 .
AFTAB J , KHAN M A , ARSHAD S , et al . Artificial intelligence based classification and prediction of medical imaging using a novel framework of inverted and self-attention deep neural network architecture [J]. Sci Rep , 2025 , 15 : 8724 . DOI: 10.1038/s41598-025-93718-7 http://dx.doi.org/10.1038/s41598-025-93718-7 .
HOFMANNINGER J , PRAYER F , PAN J , et al . Automatic lung segmentation in routine imaging is primarily a data diversity problem, not a methodology problem [J]. Eur Radiol Exp , 2020 , 4 ( 1 ): 50 . DOI: 10.1186/s41747-020-00173-2 http://dx.doi.org/10.1186/s41747-020-00173-2 .
徐辉 , 岳保荣 , 尉可道 , 等 . 《X射线计算机体层摄影装置质量控制检测规范》解读 [J]. 中国辐射卫生 , 2021 , 30 ( 6 ): 752 - 747,756 . DOI: 10.13491/j.issn.1004-714X.2021.06.018 http://dx.doi.org/10.13491/j.issn.1004-714X.2021.06.018 .
XU H , YUE B R , WEI K D , et al . Interpretation of Specification for testing of quality control in X-ray computed tomography [J]. Chin J Radiol Health , 2021 , 30 ( 6 ): 752 - 747,756 . DOI: 10.13491/j.issn.1004-714X.2021.06.018 http://dx.doi.org/10.13491/j.issn.1004-714X.2021.06.018 .
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 [J]. 结核与肺部疾病杂志 , 2024 , 5 ( 4 ): 376 - 378 . DOI: 10.19983/j.issn.2096-8493.2024022 http://dx.doi.org/10.19983/j.issn.2096-8493.2024022 .
National Health and Family Planning Commission of the People’s Republic of China . WS 288: 2017 diagnosis of tuberculosis [J]. J Tuberc Lung Dis , 2024 , 5 ( 4 ): 376 - 378 . DOI: 10.19983/j.issn.2096-8493.2024022 http://dx.doi.org/10.19983/j.issn.2096-8493.2024022 .
中国研究型医院学会感染与炎症放射学专业委员会 , 中华医学会放射学分会传染病学组 , 中国科技产业化促进会数字健康专业委员会 , 等 . 耐药肺结核影像学诊断专家共识 [J]. 新发传染病电子杂志 , 2023 , 8 ( 5 ): 63 - 74 . DOI: 10.19871/j.cnki.xfcrbzz.2023.05.013 http://dx.doi.org/10.19871/j.cnki.xfcrbzz.2023.05.013 .
Professional Committee of Infectious and Inflammatory Radiology of Chinese Research Hospital Association ; Radiology Branch of Chinese Medical Association , Group of Infectious Diseases ; China Association for the Promotion of Science and Technology Industrialization , Digital Health Professional Committee , et al . Expert consensus on imaging diagnosis of drug-resistant pulmonary tuberculosis [J]. Electron J Emerg Infect Dis , 2023 , 8 ( 5 ): 63 - 74 . DOI: 10.19871/j.cnki.xfcrbzz.2023.05.013 http://dx.doi.org/10.19871/j.cnki.xfcrbzz.2023.05.013 .
KAZEMZADEH S , YU J , JAMSHY S , et al . Deep learning detection of active pulmonary tuberculosis at chest radiography matched the clinical performance of radiologists [J]. Radiology , 2023 , 306 ( 1 ): 124 - 137 . DOI: 10.1148/radiol.212213 http://dx.doi.org/10.1148/radiol.212213 .
HWANG E J , PARK S , JIN K N , et al . Development and validation of a deep learning-based automatic detection algorithm for active pulmonary tuberculosis on chest radiographs [J]. Clin Infect Dis , 2019 , 69 ( 5 ): 739 - 747 . DOI: 10.1093/cid/ciy967 http://dx.doi.org/10.1093/cid/ciy967 .
RAJPURKAR P , O’CONNELL C , SCHECHTER A , et al . CheXaid: deep learning assistance for physician diagnosis of tuberculosis using chest x-rays in patients with HIV [J]. NPJ Digit Med , 2020 , 3 : 115 . DOI: 10.1038/s41746-020-00322-2 http://dx.doi.org/10.1038/s41746-020-00322-2 .
WANG C D , MA J C , ZHANG S , et al . Development and validation of an abnormality-derived deep-learning diagnostic system for major respiratory diseases [J]. NPJ Digit Med , 2022 , 5 ( 1 ): 124 . DOI: 10.1038/s41746-022-00648-z http://dx.doi.org/10.1038/s41746-022-00648-z .
QIN Z Z , AHMED S , SARKER M S , et al . Tuberculosis detection from chest x-rays for triaging in a high tuberculosis-burden setting: an evaluation of five artificial intelligence algorithms [J]. Lancet Digit Health , 2021 , 3 ( 9 ): e543 - e554 . DOI: 10.1016/S2589-7500(21)00116-3 http://dx.doi.org/10.1016/S2589-7500(21)00116-3 .
DE CAMARGO T F O , RIBEIRO G A S , DA SILVA M C B , et al . Clinical validation of an artificial intelligence algorithm for classifying tuberculosis and pulmonary findings in chest radiographs [J]. Front Artif Intell , 2025 , 8 : 1512910 . DOI: 10.3389/frai.2025.1512910 http://dx.doi.org/10.3389/frai.2025.1512910 .
DENG J J , ZHAO M M , LI Q Y , et al . Implementation of artificial intelligence in the histological assessment of pulmonary subsolid nodules [J]. Transl Lung Cancer Res , 2021 , 10 ( 12 ): 4574 - 4586 . DOI: 10.21037/tlcr-21-971 http://dx.doi.org/10.21037/tlcr-21-971 .
SHI C Z , SHAO Y , SHAN F , et al . Development and validation of a deep learning model for multicategory pneumonia classification on chest computed tomography: a multicenter and multireader study [J]. Quant Imaging Med Surg , 2023 , 13 ( 12 ): 8641 - 8656 . DOI: 10.21037/qims-23-1097 http://dx.doi.org/10.21037/qims-23-1097 .
BAI H X , WANG R , XIONG Z , et al . Artificial intelligence augmentation of radiologist performance in distinguishing COVID-19 from pneumonia of other origin at chest CT [J]. Radiology , 2021 , 299 ( 1 ): E225 . DOI: 10.1148/radiol.2021219004 http://dx.doi.org/10.1148/radiol.2021219004 .
李文婷 , 王丽 , 方勇 , 等 . 基于CT影像的深度学习模型在肺结核与非结核分枝杆菌肺病中的鉴别诊断价值 [J]. 中国防痨杂志 , 2024 , 46 ( 10 ): 1236 - 1242 . DOI: 10.19982/j.issn.1000-6621.20240263 http://dx.doi.org/10.19982/j.issn.1000-6621.20240263 .
LI W T , WANG L , FANG Y , et al . The value of CT-based deep learning models in differentiating nontuberculous mycobacterial lung disease from pulmonary tuberculosis [J]. Chin J Antituberc , 2024 , 46 ( 10 ): 1236 - 1242 . DOI: 10.19982/j.issn.1000-6621.20240263 http://dx.doi.org/10.19982/j.issn.1000-6621.20240263 .
刘雪艳 , 王芳 , 李春华 , 等 . 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 http://dx.doi.org/10.19982/j.issn.1000-6621.20230356 .
LIU X Y , WANG F , LI C H , et al . Construction and evaluation of a CT-based deep learning model for the auxiliary diagnosis of secondary tuberculosis [J]. Chin J Antituberc , 2024 , 46 ( 3 ): 279 - 287 . DOI: 10.19982/j.issn.1000-6621.20230356 http://dx.doi.org/10.19982/j.issn.1000-6621.20230356 .
杨佳 , 吕圣秀 , 唐光孝 , 等 . 初治与复治耐多药肺结核患者的CT表现分析 [J]. 中国防痨杂志 , 2020 , 42 ( 1 ): 38 - 43 .
YANG J , LYU S X , TANG G X , et al . Analysis of CT findings in patients with initial and retreated multidrug-resistant tuberculosis [J]. Chin J Antituberc , 2020 , 42 ( 1 ): 38 - 43 .
GAO X W , QIAN Y . Prediction of multidrug-resistant TB from CT pulmonary images based on deep learning techniques [J]. Mol Pharm , 2018 , 15 ( 10 ): 4326 - 4335 . DOI: 10.1021/acs.molpharmaceut.7b00875 http://dx.doi.org/10.1021/acs.molpharmaceut.7b00875 .
LI Y , WANG B , WEN L M , et al . Machine learning and radiomics for the prediction of multidrug resistance in cavitary pulmonary tuberculosis: a multicentre study [J]. Eur Radiol , 2023 , 33 ( 1 ): 391 - 400 . DOI: 10.1007/s00330-022-08997-9 http://dx.doi.org/10.1007/s00330-022-08997-9 .
JAEGER S , JUAREZ-ESPINOSA O H , CANDEMIR S , et al . Detecting drug-resistant tuberculosis in chest radiographs [J]. Int J Comput Assist Radiol Surg , 2018 , 13 ( 12 ): 1915 - 1925 . DOI: 10.1007/s11548-018-1857-9 http://dx.doi.org/10.1007/s11548-018-1857-9 .
LIANG S F , XU X Y , YANG Z , et al . Deep learning for precise diagnosis and subtype triage of drug-resistant tuberculosis on chest computed tomography [J]. MedComm (2020), 2024 , 5 ( 3 ): e487 . DOI: 10.1002/mco2.487 http://dx.doi.org/10.1002/mco2.487 .
0
浏览量
0
下载量
0
CSCD
关联资源
相关文章
相关作者
相关机构
京公网安备11010802024621