基于深度學習的宮頸癌異常細胞快速檢測方法

姚超; 趙基淮; 馬博淵; 李莉; 馬瑩; 班曉娟; 姜淑芳; 邵炳衡

doi:10.13374/j.issn2095-9389.2021.01.12.001

基于深度學習的宮頸癌異常細胞快速檢測方法

doi: 10.13374/j.issn2095-9389.2021.01.12.001

姚超^{1, 2, 3, 4},
趙基淮^{1, 2, 3, 4},
馬博淵^{1, 2, 3, 4, 5},
李莉^{6, 7},
馬瑩^{6, 7},
班曉娟^{1, 2, 3, 4, 5},
姜淑芳^{6, 7, ,},
邵炳衡⁸

1.
北京科技大學北京材料基因工程高精尖創新中心，北京 100083
2.
北京科技大學材料領域知識工程北京市重點實驗室，北京 100083
3.
北京科技大學人工智能研究院，北京 100083
4.
北京科技大學計算機與通信工程學院，北京 100083
5.
北京科技大學順德研究生院，佛山 528300
6.
解放軍總醫院海南醫院婦產科，三亞 572000
7.
解放軍總醫院第一醫學中心婦產科，北京 100853
8.
吉林大學口腔醫學院，長春 130021

基金項目: 海南省財政科技計劃資助項目（ZDYF2019009）；國家自然科學基金資助項目（61873299，61902022，61972028，6210020684）；中央高校基本科研業務費資助項目（FRF-TP-19-015A1，FRF-TP-20-061A1Z，FRF-TP-19-043A2，00007467）；佛山市科技創新專項資金資助項目（BK19AE034，BK20AF001，BK21BF002）

詳細信息

通訊作者:
E-mail: jsf0912@aliyun.com

中圖分類號: TP391
計量
- 文章訪問數: 1742
- HTML全文瀏覽量: 601
- PDF下載量: 226
- 被引次數: 0
出版歷程
- 收稿日期: 2021-01-12
- 網絡出版日期: 2021-06-18
- 刊出日期: 2021-09-18

Fast detection method for cervical cancer abnormal cells based on deep learning

YAO Chao^{1, 2, 3, 4},
ZHAO Ji-huai^{1, 2, 3, 4},
MA Bo-yuan^{1, 2, 3, 4, 5},
LI Li^{6, 7},
MA Ying^{6, 7},
BAN Xiao-juan^{1, 2, 3, 4, 5},
JIANG Shu-fang^{6, 7
, ,},
SHAO Bing-heng⁸

1.
Beijing Advanced Innovation Center for Materials Genome Engineering, University of Science and Technology Beijing, Beijing 100083, China
2.
Beijing Key Laboratory of Knowledge Engineering for Materials Science, University of Science and Technology Beijing, Beijing 100083, China
3.
School of Computer and Communication Engineering, University of Science and Technology Beijing, Beijing 100083, China
4.
Institute of Artificial Intelligence, University of Science and Technology Beijing, Beijing 100083, China
5.
Shunde Graduate School, University of Science and Technology Beijing, Foshan 528300, China
6.
Department of Obstetrics and Gynecology, Hainan Hospital of PLA General Hospital, Sanya 572000, China
7.
Department of Obstetrics and Gynecology, General Hospital of PLA, Beijing 100853, China
8.
School of Stomatology, Jilin University, Changchun 130021, China

More Information

Corresponding author: E-mail: jsf0912@aliyun.com

摘要

摘要: 宮頸癌是嚴重危害婦女健康的惡性腫瘤，威脅著女性的生命，而通過基于圖像處理的細胞學篩查是癌前篩查的最為廣泛的檢測方法。近年來，隨著以深度學習為代表的機器學習理論的發展，卷積神經網絡以其強有效的特征提取能力取得了圖像識別領域的革命性突破，被廣泛應用于宮頸異常細胞檢測等醫療影像分析領域。但由于病理細胞圖像具有分辨率高和尺寸大的特點，且其大多數局部區域內都不含有細胞簇，深度學習模型采用窮舉候選框的方法進行異常細胞的定位和識別時，經過窮舉候選框獲得的子圖大部分都不含有細胞簇。當子圖數量逐漸增加時，大量不含細胞簇的圖像作為目標檢測網絡輸入會使圖像分析過程存在冗余時長，嚴重減緩了超大尺寸病理圖像分析時的檢測速度。本文提出一種新的宮頸癌異常細胞檢測策略，針對使用膜式法獲得的病理細胞圖像，通過基于深度學習的圖像分類網絡首先判斷局部區域是否出現異常細胞，若出現則進一步使用單階段的目標檢測方法進行分析，從而快速對異常細胞進行精確定位和識別。實驗表明，本文提出的方法可提高一倍的宮頸癌異常細胞檢測速度。
- 宮頸癌異常細胞 /
- 病理圖像 /
- 深度學習 /
- 卷積神經網絡 /
- 目標檢測 /
- 圖像分類
Abstract: Cervical cancer is a malignant tumor that highly endangers women’s lives. Cytological screening based on image processing is the most widely used detection method for precancerous screening. Recently, with the development of machine learning theory based on deep learning, the convolutional neural network has made a revolutionary breakthrough in the field of image recognition due to its strong and effective extraction ability. In addition, it has been widely used in the field of medical image analysis such as cervical abnormal cell detection. However, due to the characteristic high resolution and large size of pathological cell images, most of its local areas do not contain cell clusters. Moreover, when the deep learning model uses the method of exhausting candidate boxes to locate and identify abnormal cells, most of the sub-images obtained do not contain cell clusters. When the number of sub-images increases gradually, a large number of images without cell clusters as input to the object detection network will make the image analysis process redundant for a long time, which drastically slows down the speed of detection of the large-scale pathological image analysis. In view of this, this paper proposed a new detection strategy for abnormal cells in cervical cancer microscopic imaging. According to the pathological cell images obtained by the membrane method, the image classification network based on deep learning was first used to determine whether there were abnormal cells in the local area. If there were abnormal cells in the local area, the single-stage object detection method was used for further pathological cell image analysis, so that the abnormal cells in the images could be quickly and accurately located and identified. Experimental results show that the proposed method can double the speed of detection of cervical cancer abnormal cells.
- cervical cancer abnormal cells /
- pathological image /
- deep learning /
- convolutional neural network /
- object detection /
- image classification

HTML全文

圖 1 本文提出的加速策略的技術路線流程圖（圖中紅色框代表異常細胞）

Figure 1. Flow chart of the proposed acceleration strategy (The red box indicates an abnormal cell)

下載: 全尺寸圖片幻燈片

圖 2 “滑動交疊裁剪”示例

Figure 2. Example of “sliding overlap clipping”

下載: 全尺寸圖片幻燈片

圖 3 YoloV5網絡結構

Figure 3. YoloV5 network structure

下載: 全尺寸圖片幻燈片

圖 4 數據集中部分細胞簇的識別示例。（a,b）正確識別結果；（c）“過檢”識別結果；（d）“漏檢”識別結果

Figure 4. Examples of the identification of some cell clusters in datasets: (a,b) correct recognition results; (c) recognition results of "over inspection"; (d) recognition results of "over inspection"

下載: 全尺寸圖片幻燈片

表 1 圖像標注情況

Table 1. Image annotation

Category	Number
ASC-US	2032
ASC-H	1156
LSIL	4387
HSIL	1389
Total	8964

下載: 導出CSV

表 2 細胞簇圖像分類實驗

Table 2. Cell cluster image classification experiment

Model	Accuracy/%	True negative rate/%	True positive rate/%	Average time consumption/s	Params/MB	Memory Cost/GB
Resnet50	89.01	96.09	86.93	0.017	22.51	4.12
Resnet101	89.62	89.39	91.46	0.027	42.50	7.85
SE-Resnext50	84.59	96.09	79.90	0.016	27.56	4.28
SE-Resnext101	82.50	91.62	79.90	0.033	48.96	8.05
Efficientnet-b4	75.71	98.88	57.29	0.027	19.43	5.12
Efficientnet-b7	83.41	98.88	68.84	0.043	66.52	25.32
Resnext50_32X4d	88.25	94.41	88.44	0.012	25.03	4.29
Resnext101_32X4d	87.20	89.94	91.46	0.025	44.18	8.03
SE-Resnet101	82.50	92.18	79.40	0.023	49.33	7.63
SE-Resnet50	85.11	88.83	86.43	0.011	28.09	3.9
Nasnet	85.37	99.44	72.36	0.038	88.75	24.04
Shufflenetv2	81.46	0	99.50	0.010	7.39	0.60
Inceptionv4	81.72	99.44	0	0.024	42.68	12.31
Xception	78.85	99.44	99.50	0.015	22.86	8.42
Densenet121	80.58	94.41	56.28	0.021	7.98	2.88

下載: 導出CSV

表 3 模型推理時間對比實驗

Table 3. Comparison experiment for model reasoning time

Single stage model	Time consumption/s	Param/MB	Double stage models	Time consumption/s	Param/MB
Faster RCNN	2775	40.1	Resnet50+Faster RCNN	1089	62.61
Cascade RCNN	2877	65.9	Resnet50+Cascade RCNN	1178	88.41
Libra RCNN	3118	41.6	Resnet50+Libra RCNN	1496	64.11
Tridentnet	4469	33.1	Resnet50+Tridentnet	2106	55.61
Foveabox	2437	36.0	Resnet50+Foveabox	1189	58.51
ATSS	3014	31.2	Resnet50+ATSS	1450	53.71
YoloV5	1386	45.7	Resnet50+YoloV5	695	68.21

下載: 導出CSV

表 4 模型識別精度對比實驗

Table 4. Comparison experiment for model recognition accuracy

Single stage model	AP50/%	Double stage models	AP50/%
Faster RCNN	70.1	Resnet50+Faster RCNN	66.8
Cascade RCNN	69.2	Resnet50+Cascade RCNN	65.7
Libra RCNN	68.3	Resnet50+Libra RCNN	67.0
Tridentnet	65.7	Resnet50+Tridentnet	59.7
Foveabox	67.3	Resnet50+Foveabox	61.9
ATSS	63.8	Resnet50+ATSS	63.5
YoloV5	75.3	Resnet50+YoloV5	70.1

下載: 導出CSV

www.77susu.com

參考文獻(44)

[1]	Ma D, Xi L. Special discussion on cervical cancer: Research Progress in epidemiology and etiology of cervical cancer. J Pract Obstet Gynecol, 2001, 17(2): 61 doi: 10.3969/j.issn.1003-6946.2001.02.001 馬丁, 奚玲. 宮頸癌專題討論——宮頸癌流行病學及病因學研究進展. 實用婦產科雜志, 2001, 17(2):61 doi: 10.3969/j.issn.1003-6946.2001.02.001
[2]	Li C C, Zhu L. The cause of cervical cancer and the status of cervical cancer vaccine. J Mod Oncol, 2018, 26(20): 3333 doi: 10.3969/j.issn.1672-4992.2018.20.040 李聰聰, 朱莉. 宮頸癌的病因及宮頸癌疫苗現狀. 現代腫瘤醫學, 2018, 26(20):3333 doi: 10.3969/j.issn.1672-4992.2018.20.040
[3]	Wu S C. Significance of gynecological census in prevention of cervical cancer. Guide China Med, 2011, 9(27): 91 doi: 10.3969/j.issn.1671-8194.2011.27.065 吳三春. 婦科普查對預防宮頸癌的意義. 中國醫藥指南, 2011, 9(27):91 doi: 10.3969/j.issn.1671-8194.2011.27.065
[4]	Zhang W H, Li N, Wu L Y. Attention should be paid on the trend of carcinoma of the cervix in young women. Zhejiang Cancer J, 2000, 6(2): 112 章文華, 李楠, 吳令英. 重視宮頸癌患者年輕化的趨勢. 浙江腫瘤, 2000, 6(2):112
[5]	Lin Y X. Clinical Pathological Features and Prognosis Analysis of Cervical Cancer [Dissertation]. Nanning: Guangxi Medical University, 2013 林泳秀. 宮頸癌的臨床病理特點與預后分析[學位論文]. 南寧: 廣西醫科大學, 2013
[6]	Wang L, Zhao W X, Zhao X L, et al. A comparative study of two methods of smear-making: Membrane-based and sedimentation in liquid-based cervical cytology. Med Pharm J Chin People’s Liberation Army, 2011, 23(4): 6 doi: 10.3969/j.issn.2095-140X.2011.04.003 王力, 趙穩興, 趙璽龍, 等. 膜式和沉降式宮頸液基細胞學制片方法的比較研究. 解放軍醫藥雜志, 2011, 23(4):6 doi: 10.3969/j.issn.2095-140X.2011.04.003
[7]	Zhao C Q, Zhou X R, Sui L, et al. Cervical Cancer Screening and Clinical Management (Cytology, Histology, Colposcopy). Beijing: Beijing Science and Technology Press, 2017 趙澄泉, 周先榮, 隋龍, 等. 宮頸癌篩查及臨床處理(細胞學、組織學和陰道鏡學). 北京: 北京科學技術出版社, 2017
[8]	Wei L H. Standardized Training Materials for Colposcopy and Cervical Cytopathology. Beijing: People’s Medical Publishing House, 2020 魏麗惠. 陰道鏡及宮頸細胞病理學規范化培訓教材. 北京: 人民衛生出版社, 2020
[9]	Wang Y Y, Wang Y, Qiao Y L, et al. The future of cervical cancer screening in China?Abandonment or conservation of cytological primary screening. Chin J Pract Gynecol Obstet, 2017, 33(3): 324 王軼英, 王悅, 喬友林, 等. 中國宮頸癌篩查未來之路——細胞學初篩的棄或守. 中國實用婦科與產科雜志, 2017, 33(3):324
[10]	LeCun Y, Bengio Y, Hinton G. Deep learning. Nature, 2015, 521(7553): 436 doi: 10.1038/nature14539
[11]	Szeliski R. Computer Vision. London: Springer London Press, 2011
[12]	Zhang Y Z. Image Engineering. 4th Ed. Beijing: Tsinghua University Press, 2018 章毓晉. 圖像工程. 4版. 北京: 清華大學出版社, 2018
[13]	Shi Y H. Study of Machine Learning Techniques and Applications in Med-Ical Image Analysis [Dissertation]. Nanjing: Nanjing University, 2013 史穎歡. 醫學圖像處理中的機器學習方法及其應用研究[學位論文]. 南京: 南京大學, 2013
[14]	Rumelhart D E, Hinton G E, Williams R J. Learning representations by back-propagating errors. Nature, 1986, 323(6088): 533 doi: 10.1038/323533a0
[15]	He K M, Zhang X Y, Ren S Q, et al. Deep residual learning for image recognition // 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, 2016: 770
[16]	Cai W B. Study on Detection Techniques of Pathological Images of Cancer Cells [Dissertation]. Taiyuan: North University of China, 2018 蔡武斌. 癌細胞病理圖像的檢測技術研究[學位論文]. 太原: 中北大學, 2018
[17]	Kermany D S, Goldbaum M, Cai W J, et al. Identifying medical diagnoses and treatable diseases by image-based deep learning. Cell, 2018, 172(5): 1122 doi: 10.1016/j.cell.2018.02.010
[18]	Cong M, Wu T, Liu D, et al. Prostate MR/TRUS image segmentation and registration methods based on supervised learning. Chin J Eng, 2020, 42(10): 1362 叢明, 吳童, 劉冬, 等. 基于監督學習的前列腺MR/TRUS圖像分割和配準方法. 工程科學學報, 2020, 42(10):1362
[19]	Zhang L, Le L, Nogues I, et al. DeepPap: deep convolutional networks for cervical cell classification. IEEE J Biomed Heal Inform, 2017, 21(6): 1633 doi: 10.1109/JBHI.2017.2705583
[20]	Wu M, Yan C B, Liu H Q, et al. Automatic classification of cervical cancer from cytological images by using convolutional neural network. Biosci Rep, 2018, 38(6): BSR20181769 doi: 10.1042/BSR20181769
[21]	Jia D Y, Li Z Y, Zhang C W. Detection of cervical cancer cells based on strong feature CNN-SVM network. Neurocomputing, 2020, 411: 112 doi: 10.1016/j.neucom.2020.06.006
[22]	Ren S Q, He K M, Girshick R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks. IEEE Trans Pattern Anal Mach Intell, 2017, 39(6): 1137 doi: 10.1109/TPAMI.2016.2577031
[23]	Chen X H. The Bethesda System for Reporting Cervical Cytology (Definitions, Criteria, and Explanatory Notes). 3rd Ed. Beijing: Science press, 2019 陳小槐. 子宮頸細胞學Bethesda報告系統(中文翻譯版, 原書第3版). 北京: 科學出版社, 2019
[24]	He Y E, Wang Y F, Lang J H, et al. The evaluation of computer cytological test with colposcopy for the diagnosis of cervical lesions. Chin J Obstet Gynecol, 1998, 33(5): 265 doi: 10.3760/j.issn:0529-567X.1998.05.002 賀又娥, 王友芳, 郎景和, 等. 計算機輔助細胞檢測系統配合陰道鏡檢查對子宮頸病變的診斷價值. 中華婦產科雜志, 1998, 33(5):265 doi: 10.3760/j.issn:0529-567X.1998.05.002
[25]	Stoler M H. Advances in cervical screening technology. Mod Pathol, 2000, 13(3): 275 doi: 10.1038/modpathol.3880048
[26]	Zhou L P. Cervical Smears Automatic Auxiliary System Related to Image Interpretation Techniques [Dissertation]. Qingdao: Ocean University of China, 2009 周立平. 宮頸涂片自動輔助判讀系統中相關圖像技術的研究[學位論文]. 青島: 中國海洋大學, 2009
[27]	Zheng K, Zhang S, Tang J Q. Application of Thinprep computer-assisted imaging system in cervical cytology. Chin J Diagn Pathol, 2015, 22(6): 364 doi: 10.3969/j.issn.1007-8096.2015.06.013 鄭珂, 張聲, 唐堅清. 計算機輔助閱片系統在宮頸細胞學篩查中的應用. 診斷病理學雜志, 2015, 22(6):364 doi: 10.3969/j.issn.1007-8096.2015.06.013
[28]	Lin T Y, Maire M, Belongie S, et al. Microsoft COCO: common objects in context // Computer Vision – ECCV 2014. Zurich, 2014: 740
[29]	Paszke A, Gross S, Massa F, et al. Pytorch: An imperative style, high-performance deep learning library[J/OL]. arXiv preprint (2019-12-3) [2021-6-24]. https://arxiv.org/abs/1912.01703
[30]	Mao X Y, Leng X F. Introduction to Opencv3 programming. Beijing: Electronics industry publishing house, 2015 毛星云, 冷雪飛. Opencv3編程入門. 北京: 電子工業出版社, 2015
[31]	Huang G, Liu Z, Pleiss G, et al. Convolutional networks with dense connectivity. IEEE Trans Pattern Anal Mach Intell, 2019, doi: 10.1109/TPAMI.2019.2918284
[32]	Hu J, Shen L, Sun G. Squeeze-and-excitation networks // 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, 2018: 7132
[33]	Tan M, Le Q. Efficientnet: Rethinking model scaling for convolutional neural networks // International Conference on Machine Learning. Long Beach, 2019: 6105
[34]	Xie S N, Girshick R, Dollár P, et al. Aggregated residual transformations for deep neural networks // 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu, 2017: 5987
[35]	Zoph B, Vasudevan V, Shlens J, et al. Learning transferable architectures for scalable image recognition // 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, 2018: 8697
[36]	Ma N N, Zhang X Y, Zheng H T, et al. ShuffleNet V2: practical guidelines for efficient CNN architecture design // Computer Vision – ECCV 2018. Munich, 2018: 122
[37]	Szegedy C, Ioffe S, Vanhoucke V, et al. Inception-v4, inception-ResNet and the impact of residual connections on learning // Proceedings of the AAAI Conference on Artificial Intelligence. San Francisco, 2017(31): 1
[38]	Chollet F. Xception: deep learning with depthwise separable convolutions // 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu, 2017: 1800
[39]	Cai Z W, Vasconcelos N. Cascade R-CNN: Delving into high quality object detection // 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, 2018: 6154
[40]	Pang J M, Chen K, Shi J P, et al. Libra R-CNN: Towards balanced learning for object detection // 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Long Beach, 2019: 821
[41]	Li Y H, Chen Y T, Wang N Y, et al. Scale-aware trident networks for object detection // 2019 IEEE/CVF International Conference on Computer Vision (ICCV). Seoul, 2019: 6053
[42]	Kong T, Sun F C, Liu H P, et al. FoveaBox: beyound anchor-based object detection. IEEE Trans Image Process, 2020, 29: 7389 doi: 10.1109/TIP.2020.3002345
[43]	Zhang S F, Chi C, Yao Y Q, et al. Bridging the gap between anchor-based and anchor-free detection via adaptive training sample selection // 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Seattle, 2020: 9756
[44]	Ultralytics. YoloV5[EB/OL]. Github(2020-10-12)[2021-06-24]. https://github.com/ultralytics/YoloV5