深度学习在胸部X线摄片筛查肺部疾病中的应用现状

doi:10.19982/j.issn.1000-6621.20240245

摘要/Abstract

摘要：

肺部疾病种类多样、危害严重,早期发现对提高患者生存率至关重要。近年来,深度学习技术在医学影像分析中取得突破性进展,为肺部疾病的早期筛查提供了新的可能。笔者回顾了近5年的相关研究,重点探讨了卷积神经网络、Transformer模型及其混合架构在胸部X线摄片图像分析中的应用。同时,也分析了多模型集成学习策略和注意力机制在提高肺部疾病精准诊断中的潜力,旨在全面梳理和分析采用深度学习技术在胸部X线摄片筛查肺部疾病中的应用现状、面临的挑战及未来发展方向。

关键词: X线透视检查, 神经网络(计算机), 肺疾病

Abstract:

There are various types of lung diseases with serious harm, and early detection of those diseases is crucial to improve the survival rate of patients. In recent years, deep learning technology has made a breakthrough progress in analysis of medical images, which providing new possibilities for early screening of lung diseases. The authors reviewed relevant studies in the past five years, focusing on the applications of Convolutional Neural Networks, Transformer models, and their hybrid architectures in chest X-ray (CXR) image analysis. Additionally, the potential of multi-model ensemble learning strategies and attention mechanisms in improving the diagnostic accuracy of lung diseases was also analyzed. This comprehensive review aims to systematically review and analyze the current application, challenges, and future directions of using deep learning technologies in chest X-ray screening for lung disease detection.

Key words: Fluoroscopy, Neural networks (computer), Lung diseases

中图分类号:

TP391.41
R81

李俊良, 刘欣, 林峙渊, 龙显荣, 江志航, 霍颖瑜. 深度学习在胸部X线摄片筛查肺部疾病中的应用现状[J]. 中国防痨杂志, 2024, 46(12): 1548-1559. doi: 10.19982/j.issn.1000-6621.20240245

Li Junliang, Liu Xin, Lin Zhiyuan, Long Xianrong, Jiang Zhihang, Huo Yingyu. The current application status of deep learning in Chest X-ray screening for lung diseases[J]. Chinese Journal of Antituberculosis, 2024, 46(12): 1548-1559. doi: 10.19982/j.issn.1000-6621.20240245

图/表 9

图1

表1

表2

图2

表3

卷积神经网络在X线图像肺部疾病筛查方法中的应用

文献	思想	数据	评估	优缺点
[31]	含有5个卷积层、池化层及2个密集层的模型	新型冠状病毒感染:3616;肺炎:4273;无肺部病变:10192;肺结核:3500	98.63%的准确率和98.35%的召回率	模型结构简单且效果优秀,但鲁棒性差,层数较浅
[32]	使用6种不同的卷积核并列进行特征提取,以极限学习机作为分类器	ChestXray14数据集;病毒性肺炎:1493;肺结核:1036	90.92%的准确率和96.93%的曲线下面积	模型捕获多尺度特征强,轻量化程度高
[33]	VGG19+4个密集层	新型冠状病毒感染:3615;肺炎:11726;密度增高病灶:6012;肺癌:20000;无肺部病变:37247;肺结核:1400	96.48%的准确率、93.75%的召回率、99.82%的曲线下面积	使用较小的卷积块作特征提取,通过增加网络深度或加入注意力的方式丰富多层次的特征信息
[34]	VGG16+3个3×3的卷积块和1个全局平均池化层	无肺部病变:13672;肿块:5603;新型冠状病毒感染:15660;结节:6201;胸腔积液:13501;肺炎:9878;纤维化:3357;肺结核:3184;密度增高病灶:7179;气胸:6870	98.89%的准确率和99.87%的特异度	使用较小的卷积块作特征提取,通过增加网络深度或加入注意力的方式丰富多层次的特征信息
[41]	VGG16中加入了水平和垂直方向的坐标注意机制	深圳数据集	92.73%的准确率和97.71%的曲线下面积	使用较小的卷积块作特征提取,通过增加网络深度或加入注意力的方式丰富多层次的特征信息
[36]	以ResNet50的4个具有残差结构的卷积块提取特征	肺炎胸部X线图像数据集;新型冠状病毒感染影像数据库	98.35%的准确率	以快捷连接的方式使模型更深、更复杂
[42]	DenseNet+由卷积构成的特征选择器+具有空间和通道编码能力的特征集成器	ChestX-ray14数据集;CheXpert数据集	84.1%和92.1%的准确率	从多个通道分别获取丰富的特征信息,更具有鲁棒性
[38]	VGG16、Inception v3和ResNet 50网络的集成	COVID-ChestXray数据集	特异度为98.54%,精确度为93.60%	充分利用每个网络的优势,但模型的复杂度和训练难度更大
[39]	MobileNetV2、InceptionV3和VGG19的集成	密度增高病灶:6914;无肺部病变:11192;新型冠状病毒感染:11582;肺炎:10036;肺结核:2984	四分类和五分类的准确度分别为87.12%和91.71%	充分利用每个网络的优势,但模型的复杂度和训练难度更大

表3

图3

表4

表5

图4

参考文献 61

[1]	冯筠, 牛怡, 杨晨希, 等. 融合改进变分自编码器与影像组学的X光片肺部疾病筛查算法. 西北大学学报(自然科学版), 2023, 53(3): 313-324. doi:10.16152/j.cnki.xdxbzr.2023-03-001.
[2]	Ai T, Yang Z, Hou H, et al. Correlation of Chest CT and RT-PCR Testing for Coronavirus Disease 2019 (COVID-19) in China: A Report of 1014 Cases. Radiology, 2020, 296(2): E32-E40. doi:10.1148/radiol.2020200642.
[3]	鲁慧民, 薛涵, 王奕龙, 等. 机器学习在影像组学分析中的应用综述. 计算机工程与应用, 2023, 59(17): 22-34. doi:10.3778/j.issn.1002-8331.2210-0435.
[4]	王瑜, 吴江, 伍东升. 卷积神经网络在职业性尘肺病影像学诊断中的应用研究进展. 生物医学工程学杂志, 2024, 41(2): 413-420. doi:10.7507/1001-5515.202309079.
[5]	Santosh KC, Allu S, Rajaraman S, et al. Advances in Deep Learning for Tuberculosis Screening using Chest X-rays: The Last 5 Years Review. J Med Syst, 2022, 46(11): 82. doi:10.1007/s10916-022-01870-8.
[6]	Liang S, Ma J, Wang G, et al. The Application of Artificial Intelligence in the Diagnosis and Drug Resistance Prediction of Pulmonary Tuberculosis. Front Med (Lausanne), 2022, 9: 935080. doi:10.3389/fmed.2022.935080.
[7]	李新, 陈帆, 郝海江, 等. 深度学习在肺炎检测中的研究综述. 桂林理工大学学报, 2020, 40(4): 859-866. doi:10.3969/j.issn.1674-9057.2020.04.027.
[8]	国家呼吸内科医疗质量控制中心, 中华医学会结核病学分会, 中国防痨协会结核病控制专业分会, 等. 综合医疗机构肺结核早期发现临床实践指南. 结核与肺部疾病杂志, 2024, 5(1): 1-14. doi:10.19982/j.issn.1000-6621.20230428.
[9]	杨红杰, 齐菲, 张红梅, 等. 肺癌合并肺结核的诊断和临床治疗现状. 结核与肺部疾病杂志, 2024, 5(3): 273-278. doi:10.19983/j.issn.2096-8493.2024054.
[10]	Ghiasi MM, Zendehboudi S. Application of decision tree-based ensemble learning in the classification of breast cancer. Comput Biol Med, 2021, 128: 104089. doi:10.1016/j.compbiomed.2020.104089.
[11]	Chung SW, Han SS, Lee JW, et al. Automated detection and classification of the proximal humerus fracture by using deep learning algorithm. Acta Orthop, 2018, 89(4): 468-473. doi:10.1080/17453674.2018.1453714. pmid: 29577791
[12]	Mayidili N, Jie M, Chuling H, et al. Artificial Intelligence Assisting the Early Detection of Active Pulmonary Tuberculosis from Chest X-Rays: A Population-Based Study. Front Mol Biosci, 2022: 9874475. doi:10.3389/fmolb.2022.874475.
[13]	Faiz KA, Tripti P, Belay T, et al. Computer-aided reading of tuberculosis chest radiography: moving the research agenda for-ward to inform policy. Eur Respir J, 2017, 50(1): 1700953. doi:10.1183/13993003.00953-2017.
[14]	Khan MA, Quasim MT, Alghamdi NS, et al. A secure framework for authentication and encryption using improved ECC for IoT-based medical sensor data. IEEE Access, 2020, 8: 52018-52027. doi:10.1109/access.2020.2980739.
[15]	Gao XW, James-Reynolds C, Currie E. Analysis of tuberculosis severity levels from CT pulmonary images based on enhanced residual deep learning architecture. Neurocomputing, 2020, 392: 233-244. doi:10.1016/j.neucom.2018.12.086.
[16]	Luján-García EJ, Yáñez-Márquez C, Villuendas-Rey Y, et al. A Transfer Learning Method for Pneumonia Classification and Visualization. Applied Sciences, 2020, 10(8): 2908. doi:10.3390/app10082908.
[17]	Morís DI, de Moura J, Novo J, et al. Unsupervised contrastive unpaired image generation approach for improving tuberculosis screening using chest X-ray images. Pattern Recognition Letters, 2022, 164: 60-66. doi:10.1016/j.patrec.2022.10.026.
[18]	Jaeger S, Candemir S, Antani S, et al. Two public chest X-ray datasets for computer-aided screening of pulmonary diseases. Quant Imaging Med Surg, 2014, 4(6): 475-477. doi:10.3978/j.issn.2223-4292.2014.11.20.
[19]	Shamrat FJM, Azam S, Karim A, et al. High-precision multiclass classification of lung disease through customized MobileNetV2 from chest X-ray images. Comput Biol Med, 2023, 155: 106646. doi:10.1016/j.compbiomed.2023.106646.
[20]	Cho K, Kim KD, Nam Y, et al. CheSS: Chest X-Ray Pre-trained Model via Self-supervised Contrastive Learning. J Digit Imaging, 2023, 36(3): 902-910. doi:10.1007/s10278-023-00782-4.
[21]	Chen H, Zhang T, Chen R, et al. A Novel COVID-19 Image Classification Method Based on the Improved Residual Network. Electronics, 2023, 12: 80. doi:10.3390/electronics12010080.
[22]	Badawi A, Elgazzar K. Detecting coronavirus from chest X-rays using transfer learning. Covid, 2021, 1(1): 403-415. doi:10.3390/covid1010034.
[23]	Rahman T, Khandakar A, Qiblawey Y, et al. Exploring the effect of image enhancement techniques on COVID-19 detection using chest X-ray images. Comput Biol Med, 2021, 132: 104319. doi:10.1016/j.compbiomed.2021.104319.
[24]	Shiraishi J. Development of a digital image database for chest radiographs with and without a lung nodule: ROC analysis on radiologists’ performance in detection of pulmonary nodules. Am J Roentgenol, 2000, 174(1): 71-74. doi:10.2214/ajr.174.1.1740071. pmid: 10628457
[25]	Nahiduzzaman M, Goni MOF, Anower MS, et al. A novel method for multivariant pneumonia classification based on hybrid CNN-PCA based feature extraction using extreme learning machine with CXR images. IEEE Access, 2021, 9: 147512-147526. doi:10.1109/access.2021.3123782.
[26]	Al-Sheikh MH, Al Dandan O, Al-Shamayleh AS, et al. Multi-class deep learning architecture for classifying lung diseases from chest X-Ray and CT images. Scientific Reports, 2023, 13(1): 19373. doi:10.1038/s41598-023-46147-3.
[27]	Li X, Shen L, Xie X, et al. Multi-resolution convolutional networks for chest X-ray radiograph based lung nodule detection. Artif Intell Med, 2020, 103: 101744. doi:10.1016/j.artmed.2019.101744.
[28]	Rajaraman S, Cohen G, Spear L, et al. DeBoNet: A deep bone suppression model ensemble to improve disease detection in chest radiographs. PLoS One, 2022, 17(3): e0265691. doi:10.1371/journal.pone.0265691.
[29]	Chowdary GJ, Kanhangad V. A Dual-Branch Network for Diagnosis of Thorax Diseases From Chest X-Rays. IEEE J Biomed Health Inform, 2022, 26(12): 6081-6092. doi:10.1109/JBHI.2022.3215694.
[30]	武卓越, 田雪琴, 侯潇芮, 等. 基于卷积神经网络的肺野分割和肺炎筛查. 西北大学学报(自然科学版), 2022, 52(4): 571-580. doi:10.16152/j.cnki.xdxbzr.2022-04-006.
[31]	Ahmed MS, Rahman A, AlGhamdi F, et al. Joint Diagnosis of Pneumonia, COVID-19, and Tuberculosis from Chest X-ray Images: A Deep Learning Approach. Diagnostics (Basel), 2023, 13(15): 2562. doi:10.3390/diagnostics13152562.
[32]	Nahiduzzaman M, Goni MOF, Hassan R, et al. Parallel CNN-ELM: A multiclass classification of chest X-ray images to identify seventeen lung diseases including COVID-19. Expert Syst Appl, 2023, 229: 120528. doi:10.1016/j.eswa.2023.120528.
[33]	Alshmrani GMM, Ni Q, Jiang R, et al. A deep learning architecture for multi-class lung diseases classification using chest X-ray (CXR) images. Alexandria Engineering Journal, 2023, 64: 923-935. doi:10.1016/j.aej.2022.10.053.
[34]	Shamrat FMJM, Azam S, Karim A, et al. LungNet22: A Fine-Tuned Model for Multiclass Classification and Prediction of Lung Disease Using X-ray Images. J Pers Med, 2022, 12(5): 680. doi:10.3390/jpm12050680.
[35]	Xu W, Fu YL, Zhu D. ResNet and its application to medical image processing: Research progress and challenges. Comput Methods Programs Biomed, 2023, 240: 107660. doi:10.1016/j.cmpb.2023.107660.
[36]	Farhan AMQ, Yang S. Automatic lung disease classification from the chest X-ray images using hybrid deep learning algorithm. Multimed Tools Appl, 2023, 82(25): 38561-38587. doi:10.1007/s11042-023-15047-z.
[37]	Huang H, Li Y, Wu R, et al. Benign-malignant classification of pulmonary nodule with deep feature optimization framework. Biomed Signal Process Control, 2022, 76: 103701. doi:10.1016/j.bspc.2022.103701.
[38]	Reddy ASK, Rao KNB, Soora NR, et al. Multi-modal fusion of deep transfer learning based COVID-19 diagnosis and classification using chest x-ray images. Multimed Tools Appl, 2023, 82(8): 12653-12677. doi:10.1007/s11042-022-13739-6.
[39]	Türk F, Kökver Y. Detection of Lung Opacity and Treatment Planning with Three-Channel Fusion CNN Model. Arab J Sci Eng, 2023, 14: 1-13. doi:10.1007/s13369-023-07843-4.
[40]	Qi Xuanhao, Zhi Min. Review of Attention Mechanisms in Image Processing. Front Comput Sci, 2024, 18(2): 345-362. doi:10.3778/j.issn.1673-9418.2305057.
[41]	Xu T, Yuan Z. Convolution neural network with coordinate attention for the automatic detection of pulmonary tuberculosis images on chest x-rays. IEEE Access, 2022, 10: 86710-86717. doi:10.1109/access.2022.3199419.
[42]	Guan Q, Huang Y, Luo Y, et al. Discriminative Feature Learning for Thorax Disease Classification in Chest X-ray Images. IEEE Trans Image Process, 2021, 30: 2476-2487. doi:10.1109/TIP.2021.3052711.
[43]	卢玲, 漆为民. 基于Transformer的脊椎CT图像分割. 中国图象图形学报, 2023, 28(11): 3618-3628. doi:10.11834/jig.221084.
[44]	Wang B, Zhang D, Tian Z. CoroTrans-CL: A novel transformer-based continual deep learning model for image recognition of coronavirus infections. Electronics, 2023, 12(4): 866. doi:10.3390/electronics12040866.
[45]	Han K, Wang Y, Chen H, et al. A survey on vision transformer. IEEE Trans Pattern Anal Mach Intell, 2022, 45(1): 87-110. doi:10.1109/TPAMI.2022.3152247.
[46]	Shome D, Kar T, Mohanty SN, et al. COVID-Transformer: Interpretable COVID-19 Detection Using Vision Transformer for Healthcare. Int J Environ Res Public Health, 2021, 18(21): 11086. doi:10.3390/ijerph182111086.
[47]	Okolo GI, Katsigiannis S, Ramzan N. IEViT: An enhanced vision transformer architecture for chest X-ray image classification. Comput Methods Programs Biomed, 2022, 226: 107141. doi:10.1016/j.cmpb.2022.107141.
[48]	Hao Y, Zhang C, Li X. DBM-ViT: A multiscale features fusion algorithm for health status detection in CXR/CT lungs images. Biomed Signal Process Control, 2024, 87: 105365. doi:10.1016/j.bspc.2023.105365.
[49]	Jiang X, Zhu Y, Cai G, et al. Mxt: A new variant of pyramid vision transformer for multi-label chest x-ray image classification. Cognitive Computation, 2022, 14(4): 1362-1377. doi:10.1007/s12559-022-10032-4.
[50]	Jiang X, Zhu Y, Liu Y, et al. TransDD: A transformer-based dual-path decoder for improving the performance of thoracic dis-eases classification using chest X-ray. Biomed Signal Process Control, 2024, 91: 105937. doi:10.1016/j.bspc.2023.105937.
[51]	An K, Zhang Y. A Self-Supervised Detail-Sensitive ViT-Based Model for COVID-19 X-ray Image Diagnosis: SDViT. Ap-plied Sciences, 2022, 13(1): 454. doi:10.3390/app13010454.
[52]	Chen Y, Feng J, Liu J, et al. Detection and classification of Lung Cancer cells using swin transformer. J Cancer Ther, 2022, 13(7): 464-475. doi:10.4236/jct.2022.13704.
[53]	Ma Y, Lv W. Identification of Pneumonia in Chest X-Ray Image Based on Transformer. Int J Antennas Propag, 2022, 2022(1): 5072666. doi:10.1155/2022/5072666.
[54]	Almalki EY, Zaffar M, Irfan M, et al. A Novel-based Swin Transfer Based Diagnosis of COVID-19 Patients. Intelligent Automation Soft Computing, 2023, 35(1): 163-180. doi:10.32604/iasc.2023.025580.
[55]	Duong LT, Le NH, Tran TB, et al. Detection of tuberculosis from chest X-ray images: Boosting the performance with vision transformer and transfer learning. Expert Syst Appl, 2021, 184: 115519. doi:10.1016/j.eswa.2021.115519.
[56]	Ukwuoma CC, Qin Z, Agbesi VK, et al. Dual_Pachi: Attention-based dual path framework with intermediate second order-pooling for Covid-19 detection from chest X-ray images. Comput Biol Med, 2022, 151(Pt A): 106324. doi:10.1016/j.compbiomed.2022.106324.
[57]	Bangalore Vijayakumar S, Chitty-Venkata KT, Arya K, et al. ConVision Benchmark: A Contemporary Framework to Benchmark CNN and ViT Models. AI, 2024, 5(3): 1132-1171. doi:10.3390/ai5030056.
[58]	Chen S, Ren S, Wang G, et al. Interpretable CNN-Multilevel Attention Transformer for Rapid Recognition of Pneumonia From Chest X-Ray Images. IEEE J Biomed Health Inform, 2024, 28(2): 753-764. doi:10.1109/JBHI.2023.3247949.
[59]	Wang T, Nie Z, Wang R, et al. PneuNet: deep learning for COVID-19 pneumonia diagnosis on chest X-ray image analysis using Vision Transformer. Med Biol Eng Comput, 2023, 61(6): 1395-1408. doi:10.1007/s11517-022-02746-2.
[60]	Ukwuoma CC, Qin Z, Heyat MBB, et al. Automated Lung-Related Pneumonia and COVID-19 Detection Based on Novel Feature Extraction Framework and Vision Transformer Approaches Using Chest X-ray Images. Bioengineering (Basel), 2022, 9(11): 709. doi:10.3390/bioengineering9110709.
[61]	Rajaraman S, Zamzmi G, Folio LR, et al. Detecting Tuberculosis-Consistent Findings in Lateral Chest X-Rays Using an Ensemble of CNNs and Vision Transformers. Front Genet, 2022, 13: 864724. doi:10.3389/fgene.2022.864724.

文献编号	数据集	图像种类(病种)	图片像素	数量
[18]	Shenzhen	包含336张肺结核和326张无肺部病变的胸部X线摄片图像	3000×3000	662
[18]	Montgomery	包含80张无肺部病变和58张肺结核的胸部X线摄片图像	4892×4020	138
[19]	ChestXray14	包含14种肺部病变图像(肺不张、实变、浸润、气胸、水肿、肺气肿、纤维化、胸腔积液、肺炎、胸膜增厚、心脏肥大、结节、肿块和疝气)	1024×1024	112120
[20]	CheXpert	包含14种类别图像(无肺部病变、纵隔扩大、心脏肥大、密度增高病灶、肺部病变、水肿、变实、肺炎、肺不张、气胸、胸腔积液、胸膜其他病变、骨折、辅助设备)	-	224316
[21]	CoronaHack	包括4种类别(无肺部病变、新型冠状病毒感染、病毒性肺炎、细菌性肺炎)	384×127、 1214×937、 4248×3480	5933
[22]	COVID-ChestXray-15k	包括3种图像(无肺部病变、肺炎、新型冠状病毒感染)	1024×1024	15000
[23]	COVQU	包含新型冠状病毒感染、密度增高病灶、病毒性肺炎、无肺部病变	1024×1024	21165
[17]	TBX11K	包括无肺部病变、异常但不属于肺结核图像	512×512	11200
[24]	JSRT	提供了100张恶性结节、54张良性结节和93张无结节的胸部X线摄片图像	2048×2048	247

方法	文献	预处理思想	数据	评估
数字图像处理
	[25]	自适应直方图均衡化处理	病毒性肺炎:1495;细菌性肺炎:2779;无肺部病变:1583	准确率为98.32%
	[26]	基于K-Lerch超越函数模型的图像增强处理	无肺部病变:417+277;肺炎:250+196;新型冠状病毒感染:237+337^a	X线摄片和CT图像准确率分别为98.60%和98.80%
预特征提取方法
	[27]	主动形状模型进行肺野分割+主成分分析对肋骨建模	肺结节:548;无肺部病变:177	平均曲线下面积为0.815
	[28]	以U型网络模型作为骨干模型的ResNet 18和EfficientNet B0的集成学习	新型冠状病毒感染:6032;无肺部病变:3136	准确率为98.20%,曲线下面积为0.998
	[29]	使用四路卷积编码的改进U型网络模型对胸部X线摄片图像进行肺部区域分割	ChestXray14	准确率为84.2%
	[30]	使用U型网络提取多尺度特征图,基于特征金字塔融合多尺度特征图,实现肺部轮廓分割	无肺部病变:8062;肺炎:5501;新型冠状病毒感染:562	准确率为91.3%

文献	思想	数据	评估	优缺点
[46]	Vision Transformer	无肺部病变:10314;新型冠状病毒感染:10819;肺炎:10702	98%的准确率和99%的曲线下面积	模型简单,易于实现,但是没有考虑局部信息的重要性
[47]	Vision Transformer+输入图像的3组2D卷积特征	无肺部病变:29477;肺炎:5852;肺结核:700;新型冠状病毒感染:20305	98.29%的平均准确率和98.95%的平均精确度	充分利用卷积的局部感受野和平移不变性,以及Transformer的全局注意力机制,对图像进行全局的语义建模和特征交互;Transformer的长程注意力机制容易导致图像块内部结构信息被破坏
[48]	Vision Transformer+2组不同卷积核大小的深度卷积	无肺部病变:10083+6893;新型冠状病毒感染:3466+7593;病毒性肺炎:1342+2618^a	X线和CT图像的准确率分别达到97.25%和98.36%	同上
[54]	Swin Transformer+2层密集层和2层Drop out层	新型冠状病毒感染:3616;密度增高病灶:6012;病毒性肺炎:1345;无肺部病变:10192	98%的召回率和96%的准确率	同上
[50]	结合图像的多头注意力与标签注意力的双路径解码器	ChestXray14数据集	83.1%的曲线下面积	特征图和标签向量进行信息交互,但是模型复杂、训练难度较高
[51]	Transformer图像编码+细节校正路径从X和Y像素方向的梯度信息+MagNet根据训练集分布调整测试集分布	Synthesis-covid-cxr数据集	95.23%的准确率	能够根据数据特点自适应地调节测试数据分布,聚焦更重要的内容
[49]	pyramid vision Transformer变体模型	ChestXray14数据集	83.0%的曲线下面积	位置编码的方式取代位置嵌入,减少注意力的计算量

文献	思想	数据	评估	优缺点
[55]	EfficientNet+Vision Transformer	蒙哥马利数据集、深圳数据集、新型冠状病毒感染影像数据库	97.51%的准确率和98%的召回率	优点:能够捕获多尺度上下文信息
[56]	L和AB坐标+Inception V3+Transformer	新型冠状病毒感染影像数据库、COVID-ChestX-ray-15k数据集	分别获得96.66%和97.87%的准确率	优点:能够捕获多尺度上下文信息
[57]	DenseNet与MaxViT组合成双通道模型	COVID-QU-Ex数据集	95.61%的准确率和96%的召回率	优点:减少提取特征时信息的损失,增强有用的信息,抑制不重要的信息;缺点:模型参数量大
[58]	ResNet101+多级自注意力的Transformer	无肺部病变:10192;新型冠状病毒感染:3615;密度增高病灶:6023;病毒性肺炎:1345	96.54%的准确率	优点:减少提取特征时信息的损失,增强有用的信息,抑制不重要的信息;缺点:模型参数量大
[59]	ResNet18作特征提取,提取的特征作为单独patch+Vision Transformer	肺炎:11263;新型冠状病毒感染:11956;无肺部病变:10701	94.96%的准确率	优点:特征图不是裁切成多个patch,而是对整个特征图计算注意力,直接理解全图
[60]	VGG、Googlenet、Densenet集成学习+Vision Transformer	新型冠状病毒感染影像数据库、COVID-ChestX-ray-15k数据集	分别获得98.00%和97.4%的平均精确度	优点:可有效获得不同层次之间由低阶到高阶的互补信息;缺点:特征图在多个级别上密集级联,易导致高计算资源需求
[61]	DenseNet121、VGG16、Effi-cientNet V2集成学习+Vision Transformer	肺结核:57087;无肺部病变:18946	90.57%的准确度	优点:可有效获得不同层次之间由低阶到高阶的互补信息;缺点:特征图在多个级别上密集级联,易导致高计算资源需求