基于全局-局部注意力机制和YOLOv5的宫颈细胞图像异常检测模型

doi:10.12122/j.issn.1673-4254.2024.07.01

摘要/Abstract

摘要：

目标建立一种新的基于全局-局部注意机制和YOLOv5的宫颈病变细胞检测模型（Trans-YOLOv5），为准确、高效地分析宫颈细胞学图像并做出诊断提供帮助。方法使用共含有7410张宫颈细胞学图像且均包含对应真实标签的公开数据集。采用结合了数据扩增方式与标签平滑等技巧的YOLOv5网络结构实现对宫颈病变细胞的多分类检测。在YOLOv5骨干网络引用CBT3以增强深层全局信息提取能力，设计ADH检测头提高检测头解耦后定位分支对纹理特征的结合能力，从而实现全局-局部注意机制的融合。结果实验结果表明Trans-YOLOv5优于目前最先进的方法。mAP和AR分别达到65.9%和53.3%，消融实验结果验证了Trans-YOLOv5各组成部分的有效性。结论本文发挥不同注意力机制分别在全局特征与局部特征提取能力的差异，提升YOLOv5对宫颈细胞图像中异常细胞的检测精度，展现了其在自动化辅助宫颈癌筛查工作量的巨大潜力。

关键词: 宫颈细胞图像异常检测, YOLOv5, 图像处理, 全局和局部特征融合

Abstract:

The development of various models for automated images screening has significantly enhanced the efficiency and accuracy of cervical cytology image analysis. Single-stage target detection models are capable of fast detection of abnormalities in cervical cytology, but an accurate diagnosis of abnormal cells not only relies on identification of a single cell itself, but also involves the comparison with the surrounding cells. Herein we present the Trans-YOLOv5 model, an automated abnormal cell detection model based on the YOLOv5 model incorporating the global-local attention mechanism to allow efficient multiclassification detection of abnormal cells in cervical cytology images. The experimental results using a large cervical cytology image dataset demonstrated the efficiency and accuracy of this model in comparison with the state-of-the-art methods, with a mAP reaching 65.9% and an AR reaching 53.3%, showing a great potential of this model in automated cervical cancer screening based on cervical cytology images.

Key words: cervical cancer screening, YOLOv5, image processing, Transformer

胡雯然, 傅蓉. 基于全局-局部注意力机制和YOLOv5的宫颈细胞图像异常检测模型[J]. 南方医科大学学报, 2024, 44(7): 1217-1226.

Wenran HU, Rong FU. Trans-YOLOv5: a YOLOv5-based prior transformer network model for automated detection of abnormal cells or clumps in cervical cytology images[J]. Journal of Southern Medical University, 2024, 44(7): 1217-1226.

图/表 12

参考文献 32

1	Bray F, Laversanne M, Sung H, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries［J］. CA Cancer J Clin, 2024, 74(3): 229-63. DOI: 10.3322/caac.21834
2	Solomon D, Davey D, Kurman R, et al. The 2001 Bethesda System: terminology for reporting results of cervical cytology ［J］. JAMA, 2002, 287(16): 2114-9.
3	William W, Ware A, Basaza-Ejiri AH, et al. A pap-smear analysis tool (PAT) for detection of cervical cancer from pap-smear images ［J］. Biomedical engineering online, 2019, 18: 1-22. DOI: 10.1186/s12938-019-0634-5
4	王嘉旭, 薛鹏, 江宇, 等. 人工智能在宫颈癌筛查中的应用研究进展［J］. 中国肿瘤临床, 2021, 48(09): 468-71. DOI: 10.3969/j.issn.1000-8179.2021.09.056
5	Nayar R, Wilbur DC. The Bethesda system for reporting cervical cytology: definitions, criteria, and explanatory notes ［M］. Springer, 2015. 103-285. DOI: 10.1007/978-3-319-11074-5
6	Chankong T, Theera-Umpon N, Auephanwiriyakul S. Automatic cervical cell segmentation and classification in Pap smears ［J］. Methods Programs Biomed, 2014, 113(2): 539-56. DOI: 10.1016/j.cmpb.2013.12.012
7	Bora K, Chowdhury M, Mahanta L B, et al. Automated classification of Pap smear images to detect cervical dysplasia ［J］. Comput Methods Programs Biomed, 2017, 138: 31-47. DOI: 10.1016/j.cmpb.2016.10.001
8	Arya M, Mittal N, Singh G. Texture-based feature extraction of smear images for the detection of cervical cancer ［J］. IET Comput Vision, 2018, 12(8): 1049-59. DOI: 10.1049/iet-cvi.2018.5349
9	Wang P, Wang L, Li Y, et al. Automatic cell nuclei segmentation and classification of cervical Pap smear images ［J］. Biomed Signal Process Control, 2019, 48: 93-103. DOI: 10.1016/j.bspc.2018.09.008
10	Zhang L, Lu L, Nogues I, et al. DeepPap: deep convolutional networks for cervical cell classification ［J］. IEEE J Biomed Health Inform, 2017, 21(6): 1633-43. DOI: 10.1109/jbhi.2017.2705583
11	Taha B, Dias J, Werghi N. Classification of cervical-cancer using pap-smear images: a convolutional neural network approach ［C］. Medical Image Understanding and Analysis: 21st Annual Conference, MIUA 2017, Edinburgh, UK, July 11-13, 2017, Proceedings 21. Springer International Publishing, 2017: 261-72. DOI: 10.1007/978-3-319-60964-5_23
12	Wieslander H, Forslid G, Bengtsson E, et al. Deep convolutional neural networks for detecting cellular changes due to malignancy［C］. Proceedings of the IEEE International Conference on Computer Vision Workshops. 2017: 82-9. DOI: 10.1109/iccvw.2017.18
13	Jiang P, Li X, Shen H, et al. A systematic review of deep learning-based cervical cytology screening: from cell identification to whole slide image analysis［J］. Artif Intell Rev, 2023, 56(): 2687-758. DOI: 10.1007/s10462-023-10588-z
14	Ren S, He K, Girshick R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks［J］. IEEE Trans Pattern Anal Mach Intell, 2016, 39(6): 1137-49. DOI: 10.1109/tpami.2016.2577031
15	Liang Y, Tang Z, Yan M, et al. Comparison detector for cervical cell/clumps detection in the limited data scenario［J］. Neurocomputing, 2021, 437: 195-205. DOI: 10.1016/j.neucom.2021.01.006
16	Cao L, Yang J, Rong Z, et al. A novel attention-guided convolutional network for the detection of abnormal cervical cells in cervical cancer screening［J］. Med Image Anal, 2021, 73: 102197. DOI: 10.1016/j.media.2021.102197
17	Chen T, Zheng W, Ying H, et al. A task decomposing and cell comparing method for cervical lesion cell detection［J］. IEEE Trans Med Imaging, 2022, 41(9): 2432-42. DOI: 10.1109/tmi.2022.3163171
18	Liang Y, Feng S, Liu Q, et al. Exploring contextual relationships for cervical abnormal cell detection［J］. IEEE J Biomed Health Inform, 2023, 27(8): 4086-97. DOI: 10.1109/jbhi.2023.3276919
19	Cai Z, Vasconcelos N. Cascade R-CNN: Delving into high quality object detection［C］//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018: 6154-62. DOI: 10.1109/cvpr.2018.00644
20	He K, Gkioxari G, Doll´ar P, et al. Mask R-CNN［C］. Proceedings of the IEEE International Conference on Computer Vision. 2017: 2961-9. DOI: 10.1109/iccv.2017.322
21	Yi L, Lei Y, Fan Z, et al. Automatic detection of cervical cells using dense-cascade R-CNN［C］. Pattern Recognition and Computer Vision: Third Chinese Conference, PRCV 2020, 2020, Proceedings, Part II 3. Springer International Publishing, 2020: 602-13. DOI: 10.1007/978-3-030-60639-8_50
22	Ma B, Zhang J, Cao F, et al. MACD R-CNN: an abnormal cell nucleus detection method［J］. IEEE Access, 2020, 8: 166658-69. DOI: 10.1109/access.2020.3020123
23	Redmon J, Divvala S, Girshick R, et al. You only look once: unified, real-time object detection［C］. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2016: 779-88. DOI: 10.1109/cvpr.2016.91
24	Xiang, Y, Sun W, Pan C, et al. A novel automation-assisted cervical cancer reading method based on convolutional neural network［J］. Biocybernet Biomed Engin, 2020, 40(2): 611-23. DOI: 10.1016/j.bbe.2020.01.016
25	Liang Y, Pan C, Sun W, et al. Global context-aware cervical cell detection with soft scale anchor matching［J］. Comput Methods Programs Biomed, 2021, 204: 106061. DOI: 10.1016/j.cmpb.2021.106061
26	Jia D, He Z, Zhang C, et al. Detection of cervical cancer cells in complex situation based on improved YOLOv3 network［C］. Multimed Tools Appl, 2022, 81(6): 8939-61. DOI: 10.1007/s11042-022-11954-9
27	Fei M, Zhang X, Cao M, et al. Robust cervical abnormal cell detection via distillation from local-scale consistency refinement ［C］//International Conference on Medical Image Computing and Computer-Assisted Intervention. Cham: Springer Nature Switzerland, 2023: 652-61. DOI: 10.1007/978-3-031-43987-2_63
28	Lin T Y, Goyal P, Girshick R, et al. Focal loss for dense object detection［C］//Proceedings of the IEEE International Conference on Computer Vision. 2017: 2980-2988. DOI: 10.1109/iccv.2017.324
29	Srinivas A, Lin T Y, Parmar N, et al. Bottleneck transformers for visualrecognition［C］//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021: 16519-16529. DOI: 10.1109/cvpr46437.2021.01625
30	Woo S, Park J, Lee J Y, et al. Cbam: Convolutional block attention module［C］//Proceedings of the European Conference on Computer Vision (ECCV). 2018: 3-19. DOI: 10.1007/978-3-030-01234-2_1
31	Duan S, Zhang M, Qiu S, et al. Tunnel lining crack detection model based on improved YOLOv5［J］. Tunnel Underground Space Technol, 2024, 147: 105713. DOI: 10.1016/j.tust.2024.105713
32	Song C Y, Zhang F, Li J S, et al. Detection of maize tassels for UAV remote sensing image with an improved YOLOX model［J］. J Integr Agric, 2023, 22(6): 1671-83. DOI: 10.1016/j.jia.2022.09.021

Dataset	ascus	asch	lsil	hsil	scc	agc	trich	cand	flora	herps	actin	Total	Images
Training	1614	3403	1220	23211	1759	4404	4413	288	104	235	119	40770	5922
Validation	166	470	196	2907	213	523	542	40	22	31	17	5127	744
Test	213	404	166	2803	228	662	477	26	24	36	18	5057	744
Total	1993	4277	1582	28921	2200	5589	5432	354	150	302	154	50954	7410

Dataset	ascus	asch	lsil	hsil	scc	agc	trich	cand	flora	herps	actin	Total	Images
Training	1614	3403	1220	23211	1759	4404	4413	288	104	235	119	40770	5922
Validation	166	470	196	2907	213	523	542	40	22	31	17	5127	744
Test	213	404	166	2803	228	662	477	26	24	36	18	5057	744
Total	1993	4277	1582	28921	2200	5589	5432	354	150	302	154	50954	7410

Models	Model type	mAP(%)	AR(%)	Params(M)	GFLOPs	FPS(img/s)
Faster RCNN ^[14]	Two-stage	57.6	48.1	41.5	215.8	18.6
Cascade R-CNN^[19]	Two-stage	58.1	47.8	69.2	243.5	15.9
RetinaNet^[28]	One-stage	46.2	49.1	37.7	250.3	16.3
YOLOv3	One-stage	28.8	26.6	40.6	110.7	7.4
YOLOv5l	One-stage	62.6	52.2	46.2	108.4	46.9
YOLOv8l	One-stage	60.7	43.6	43.6	165.4	78.7
AttFPN^[16]	Two-stage	37.1	40.4	41.2	192.4	14.6
Comparison Detector ^[15]	Two-stage	45.9	63.5	-	-	-
YOLOv3+SSAM^[25]	One-stage	33.2	31.9	61.6	132.8	6.2
CR4CACD^[18]	Two-stage	60.5	45.2	43.1	156.9	11.5
Trans-YOLOv5	One-stage	64.8	53.3	49.8	125.3	29.1

Models	Model type	mAP(%)	AR(%)	Params(M)	GFLOPs	FPS(img/s)
Faster RCNN ^[14]	Two-stage	57.6	48.1	41.5	215.8	18.6
Cascade R-CNN^[19]	Two-stage	58.1	47.8	69.2	243.5	15.9
RetinaNet^[28]	One-stage	46.2	49.1	37.7	250.3	16.3
YOLOv3	One-stage	28.8	26.6	40.6	110.7	7.4
YOLOv5l	One-stage	62.6	52.2	46.2	108.4	46.9
YOLOv8l	One-stage	60.7	43.6	43.6	165.4	78.7
AttFPN^[16]	Two-stage	37.1	40.4	41.2	192.4	14.6
Comparison Detector ^[15]	Two-stage	45.9	63.5	-	-	-
YOLOv3+SSAM^[25]	One-stage	33.2	31.9	61.6	132.8	6.2
CR4CACD^[18]	Two-stage	60.5	45.2	43.1	156.9	11.5
Trans-YOLOv5	One-stage	64.8	53.3	49.8	125.3	29.1

Model	ascus	asch	lsil	hsil	scc	agc	trich	cand	flora	herps	actin	mAP@50	mAP(*)
Faster RCNN ^[14]	42.9	21.1	58.9	52.7	28.8	66.6	65.3	68.3	74.4	80.7	65.3	57.6	48.1
Cascade R-CNN ^[19]	45.7	19.6	58.5	51.2	29.0	67.3	63.6	71.7	82.6	78.4	71.9	58.1	47.8
RetinaNett ^[28]	36.0	14.2	57.3	47.3	3.0	60.4	61.1	68.7	22.6	74.5	62.7	46.2	49.1
YOLOv3 ^[32]	32.5	2.8	39.8	42.8	5.8	59.7	36.2	26.4	0.0	46.3	25.0	28.9	30.2
YOLOv5l	49.5	24.6	56.4	53.2	33.1	70.6	67.9	82.8	86.9	86.5	77.6	62.6	47.6
YOLOv8l	49.9	28.3	64.5	56.7	38.4	70.9	63.6	81.3	73.5	74.3	66.6	60.7	51.8
AttFPN^[16]	26.5	19.0	32.1	45.8	26.2	55.9	53.1	32.7	25.7	55.8	37.1	37.3	35.8
Comparison Detector ^[15]	27.4	6.7	41.7	40.1	21.8	54.5	45.0	65.5	63.5	68.1	70.5	45.9	33.0
YOLOv3+SSAM^[25]	28.6	10.7	35.6	39.2	17.6	52.4	47.3	19.9	28.2	56.8	35.3	33.8	31.1
CR4CACD ^[18]	54.3	27.1	53.1	56.5	36.8	71.1	68.8	65.5	86.4	84.1	62.1	60.5	48.9
Trans-YOLOv5	55.5	32.5	64.1	60.1	36.5	74.5	67.3	87.9	78.3	88.1	79.8	65.9	53.5