基于深度学习乳腺X线摄影钙化识别分类模型的临床应用价值

doi:10.11871/jfdc.issn.2096-742X.2024.02.007

数据与计算发展前沿 ›› 2024, Vol. 6 ›› Issue (2): 68-79.

CSTR: 32002.14.jfdc.CN10-1649/TP.2024.02.007

doi: 10.11871/jfdc.issn.2096-742X.2024.02.007

• 专刊:智慧医疗前沿与进展（上） • 上一篇下一篇

基于深度学习乳腺X线摄影钙化识别分类模型的临床应用价值

袁家琳^1,²(),欧阳汝珊³,戴懿⁴,赖小慧⁵,马捷^1,^2,^*(),龚静山^1,²

1.暨南大学，第二临床医学院，广东深圳 518020
2.深圳市人民医院，放射科，广东深圳 518020
3.中山大学第八附属医院（深圳福田），放射科，广东深圳 518033
4.北京大学深圳医院，放射科，广东深圳 518036
5.深圳市罗湖区人民医院，放射科，广东深圳 518000

收稿日期:2023-10-23 出版日期:2024-04-20 发布日期:2024-04-26
通讯作者: *马捷（E-mail: cjr.majie@vip.163.com）
作者简介:袁家琳，深圳市人民医院放射科主治医师，暨南大学在读博士生。研究方向为深度学习在影像检查图像中的应用，主要擅长影像图像后处理，弥散成像，结构网络的应用。
本文中负责实验实施与统计分析、结果验证、论文撰写。
YUAN Jialin is a radiologist at Shenzhen People's Hospital and a doctoral student at Jinan University. Her research focuses on the application of deep learning in medical imaging, with expertise in image post-processing, diffusion imaging, and the application of structural networks.
In this paper, she is responsible for project implementation, statistical analysis, result validation, and paper writing.
E-mail: yuan.jialin@szhospital.com|马捷，深圳市人民医院分院放射科主任，主任医师，教授，硕士生导师，长期致力于影像诊断的临床及基础研究工作，擅长乳腺影像诊断及微创活检，美国哈佛大学布列根妇女医院访问学者。曾获得广东省科技进步奖2项，深圳市科技进步奖4项，发表学术论文60余篇，已出版著作3部，获得成果转换三项。
本文中负责研究规划、项目指导与论文修改。
MA Jie is the Director of the Radiology Department at Shenzhen People's Hospital (Longhua Branch), Chief Physician, Professor, and Master's Supervisor, and a visiting scholar at Harvard University's Brigham and Women's Hospital. Her expertise lies in breast diagnostic imaging. She is awarded for scientific and technological progress. She has published over 60 academic papers and authored three books. Furthermore, She has three technology transfer projects.
In this paper, she is responsible for research planning, project guidance, and paper revisions.
E-mail: cjr.majie@vip.163.com
基金资助:
国家自然科学基金面上项目“面向数据不确定性的多模态医学影像分析理论和方法”(62276121);广东省医学科研基金“基于深度学习的乳腺X线摄影与自然语言处理技术对乳腺内潜在恶性病变的分层评估研究”(A2024506);深圳市科技创新委员会国际科技自主合作项目“基于多模态深度学习算法对乳腺癌筛查和诊断评估系统研究”(GJHZ20220913142613025)

Assessing the Clinical Utility of a Deep Learning-Based Model for Calcification Recognition and Classification in Mammograms

YUAN Jialin^1,²(),OUYANG Rushan³,DAI Yi⁴,LAI Xiaohui⁵,MA Jie^1,^2,^*(),GONG Jingshan^1,²

1. The Second Clinical Medical College, Jinan University, Shenzhen, Guangdong 518020, China
2. Department of Radiology, Shenzhen People's Hospital, Shenzhen, Guangdong 518020, China
3. Department of Radiology, the Eighth Affiliated Hospital, Sun Yat-sen University, Shenzhen, Guangdong 518033, China
4. Department of Radiology, Peking University Shenzhen Hospital, Shenzhen, Guangdong 518036, China
5. Department of Radiology, Shenzhen Luohu People's Hospital, Shenzhen, Guangdong 518000, China

Received:2023-10-23 Online:2024-04-20 Published:2024-04-26

摘要/Abstract

摘要：

【目的】引入基于深度学习乳腺X线摄影钙化识别及分类模型，探讨深度学习技术对钙化灶的准确识别、分类和临床应用价值。【方法】采用多中心乳腺X线检查数据，分别由高-初级诊断医生及两名初级诊断医生采用不结合及结合深度学习模型进行病灶评估，评价其诊断效能。【结果】引入深度学习模型识别钙化灶能力与高-初级诊断医生及两名初级诊断医生识别钙化灶能力相仿（漏检率分别为0.81%vs.0.65%，1.14%vs.1.63%，P>0.05），深度学习模型能够有效帮助高-初级诊断医生(灵敏度0.926，AUC0.81，P=0.014)及两名初级诊断医生（灵敏度0.896，AUC0.79，P=0.049）检出可疑恶性钙化灶，特别是在良性病变中的准确率提升作用明显。【局限】仍需更多前瞻性多中心数据验证模型稳健性，也需引入不同深度学习模型比较其临床应用价值。【结论】深度学习模型有助于乳腺X线摄影钙化识别及分类评估，有助于乳腺癌大规模筛查背景下提供辅助诊断及临床策略支持。

关键词: 乳腺病变, 乳腺X线摄影术, 钙化识别, 深度学习

Abstract:

[Objective] This article is to assess the clinical application value of a deep learning-based model for recognizing and classifying mammography calcifications. [Methods] Multicenter mammography data were employed, with lesion assessments conducted by both senior-junior radiologists and two junior radiologists. The deep learning-based model was used in both standalone and combined approaches. Diagnostic performance was then evaluated. [Results] The introduction of the deep learning model demonstrates comparable capabilities to senior-junior radiologists and two junior radiologists (miss rates: 0.81% vs. 0.65%, 1.14% vs. 1.63%, P>0.05). The deep learning model effectively assists senior-junior radiologists (sensitivity 0.926, AUC 0.81, P=0.014) and two junior radiologists (sensitivity 0.896, AUC 0.79, P=0.049) in detecting suspicious calcifications, especially in benign lesions. [Limitations] The study requires more prospective multicenter data and different deep learning models to compare their clinical utility. [Conclusions] Deep learning frameworks offer valuable support for mammography calcification recognition and classification, providing rapid assistance for diagnosis and clinical strategy support.

Key words: breast lesions, mammography, calcification recognition, deep learning

袁家琳, 欧阳汝珊, 戴懿, 赖小慧, 马捷, 龚静山. 基于深度学习乳腺X线摄影钙化识别分类模型的临床应用价值[J]. 数据与计算发展前沿, 2024, 6(2): 68-79.

YUAN Jialin, OUYANG Rushan, DAI Yi, LAI Xiaohui, MA Jie, GONG Jingshan. Assessing the Clinical Utility of a Deep Learning-Based Model for Calcification Recognition and Classification in Mammograms[J]. Frontiers of Data and Computing, 2024, 6(2): 68-79, https://cstr.cn/32002.14.jfdc.CN10-1649/TP.2024.02.007.

图/表 10

图1

表1

图2

图3

图4

表2

表3

图5

图6

表4

参考文献 30

[1]	XIA C, DONG X, LI H, et al. Cancer statistics in China and United States, 2022: profiles, trends, and determinants[J/OL]. Chinese Medical Journal, 2022, 135(5): 584-590. https://doi.org/10.1097/CM9.0000000000002108.
[2]	SUNG H, FERLAY J, SIEGEL R L, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries[J/OL]. CA: a cancer journal for clinicians, 2021, 71(3): 209-249. https://doi.org/10.3322/caac.21660.
[3]	HOOSHMAND S, REED W M, SULEIMAN M E, et al. A review of screening mammography: The benefits and radiation risks put into perspective[J/OL]. Journal of Medical Imaging and Radiation Sciences, 2022, 53(1): 147-158. https://doi.org/10.1016/j.jmir.2021.12.002.
[4]	ZHANG M, LIN Q, SU X H, et al. Breast ductal carcinoma in situ with micro-invasion versus ductal carcinoma in situ: a comparative analysis of clinicopathological and mammographic findings[J/OL]. Clinical Radiology, 2021, 76(10): 787.e1-787.e7. https://doi.org/10.1016/j.crad.2021.04.011.
[5]	AZAM S, ERIKSSON M, SJÖLANDER A, et al. Mammographic microcalcifications and risk of breast cancer[J/OL]. British Journal of Cancer, 2021, 125(5): 759-765. https://doi.org/10.1038/s41416-021-01459-x.
[6]	MADANI M, BEHZADI M M, NABAVI S. The Role of Deep Learning in Advancing Breast Cancer Detection Using Different Imaging Modalities: A Systematic Review[J/OL]. Cancers, 2022, 14(21): 5334. https://doi.org/10.3390/cancers14215334.
[7]	MRIDHA M F, HAMID Md A, MONOWAR M M, et al. A Comprehensive Survey on Deep-Learning-Based Breast Cancer Diagnosis[J/OL]. Cancers, 2021, 13(23): 6116. https://doi.org/10.3390/cancers13236116.
[8]	CHEN Q Q, LIN S T, YE J Y, et al. Diagnostic value of mammography density of breast masses by using deep learning[J/OL]. Frontiers in Oncology, 2023, 13: 1110657. https://doi.org/10.3389/fonc.2023.1110657.
[9]	CHOI W J, AN J K, WOO J J, et al. Comparison of Diagnostic Performance in Mammography Assessment: Radiologist with Reference to Clinical Information Versus Standalone Artificial Intelligence Detection[J/OL]. Diagnostics, 2023, 13(1): 117. https://doi.org/10.3390/diagnostics13010117.
[10]	马捷, 左敏, 孙国平, 等. 乳腺癌钼靶X线表现与c-erbB-2癌基因相关性[J/OL]. 放射学实践, 20150228(03)[2023-08-12]. http://www.fsxsj.net/ch/reader/view_abstract.aspx?file_no=20060314&flag=1.
[11]	FARROKH D, BOLOURSAZ S, HOMAI F. Relationship among mammographic findings with histopathologic type of breast cancer and human epidermal growth factor receptor 2 (HER2) in young women[J/OL]. Electronic Physician, 2017, 9(5): 4300-4305. https://doi.org/10.19082/4300.
[12]	HOSSAIN M S. Microc alcification Segmentation Using Modified U-net Segmentation Network from Mammogram Images[J/OL]. Journal of King Saud University-Computer and Information Sciences, 2022, 34(2): 86-94. https://doi.org/10.1016/j.jksuci.2019.10.014.
[13]	GUO X, O’NEILL W C, VEY B, et al. SCU-Net: A deep learning method for segmentation and quantification of breast arterial calcifications on mammograms[J/OL]. Medical Physics, 2021, 48(10): 5851-5861. https://doi.org/10.1002/mp.15017.
[14]	PESAPANE F, TRENTIN C, FERRARI F, et al. Deep learning performance for detection and classification of microcalcifications on mammography[J/OL]. European Radiology Experimental, 2023, 7(1): 69. https://doi.org/10.1186/s41747-023-00384-3.
[15]	MAKEEV A, RODAL G, GHAMMRAOUI B, et al. Exploring CNN potential in discriminating benign and malignant calcifications in conventional and dual-energy FFDM: simulations and experimental observations[J/OL]. Journal of Medical Imaging, 2021, 8(3): 033501. https://doi.org/10.1117/1.JMI.8.3.033501.
[16]	CHEN J L, CHENG L H, WANG J, et al. A YOLO-based AI system for classifying calcifications on spot magnification mammograms[J/OL]. BioMedical Engineering OnLine, 2023, 22(1): 54. https://doi.org/10.1186/s12938-023-01115-w.
[17]	SONGSAENG C, WOODTICHARTPREECHA P, CHAICHULEE S. Multi-Scale Convolutional Neural Networks for Classification of Digital Mammograms With Breast Calcifications[J/OL]. IEEE Access, 2021, 9: 114741-114753. https://doi.org/10.1109/ACCESS.2021.3104627. doi: 10.1109/ACCESS.2021.3104627
[18]	ALRUWAILI M, GOUDA W. Automated Breast Cancer Detection Models Based on Transfer Learning[J/OL]. Sensors (Basel, Switzerland), 2022, 22(3): 876. https://doi.org/10.3390/s22030876.
[19]	WANG J, SUN H, JIANG K, et al. CAPNet: Context attention pyramid network for computer-aided detection of microcalcification clusters in digital breast tomosynthesis[J/OL]. Computer Methods and Programs in Biomedicine, 2023, 242: 107831. https://doi.org/10.1016/j.cmpb.2023.107831.
[20]	UEDA D, YAMAMOTO A, ONODA N, et al. Development and validation of a deep learning model for detection of breast cancers in mammography from multi-institutional datasets[J/OL]. PLOS ONE, 2022, 17(3): e0265751. https://doi.org/10.1371/journal.pone.0265751.
[21]	M C S, M V. Hybrid Transfer Learning of Mammogram Images for Screening of Micro-Calcifications[J/OL]. International Journal of Electrical and Electronics Engineering, 2022, 9(8): 40-47. https://doi.org/10.14445/23488379/IJEEE-V9I8P105.
[22]	LIN X, WU S, LI L, et al. Automatic mammographic breast density classification in Chinese women: clinical validation of a deep learning model[J/OL]. Acta Radiologica, 2023: 028418512311520. https://doi.org/10.1177/02841851231152097.
[23]	LIU H, CHEN Y, ZHANG Y, et al. A deep learning model integrating mammography and clinical factors facilitates the malignancy prediction of BI-RADS 4 microcalcifications in breast cancer screening[J/OL]. European Radiology, 2021, 31(8): 5902-5912. https://doi.org/10.1007/s00330-020-07659-y.
[24]	中华医学会影像技术分会, 中华医学会放射学分会. 乳腺影像检查技术专家共识[J]. 中华放射学杂志, 2016, 50(8): 561-565.
[25]	ZHANG Y, TANG Y, CAO Z, et al. BI-RADS Classification of Calcification on Mammograms[M/OL]//DE BRUIJNE M, CATTIN P C, COTIN S, et al. Medical Image Computing and Computer Assisted Intervention-MICCAI 2021: 卷12907. Cham: Springer International Publishing, 2021: 119-128[2023-10-13]. https://link.springer.com/10.1007/978-3-030-87234-2_12.
[26]	CAO Z, YANG Z, ZHUO X, et al. DeepLIMa: Deep Learning Based Lesion Identification in Mammograms[C/OL]// 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW), 2019: 362-370. https://doi.org/10.1109/ICCVW.2019.00047.
[27]	KERSCHKE L, WEIGEL S, RODRIGUEZ-RUIZ A, et al. Using deep learning to assist readers during the arbitration process: a lesion-based retrospective evaluation of breast cancer screening performance[J/OL]. European Radiology, 2022, 32(2): 842-852. https://doi.org/10.1007/s00330-021-08217-w.
[28]	YANG Z, CAO Z, ZHANG Y, et al. MommiNet-v2: Mammographic multi-view mass identification networks[J/OL]. Medical Image Analysis, 2021, 73: 102204. https://doi.org/10.1016/j.media.2021.102204.
[29]	KANG D, GWEON H M, EUN N L, et al. A convolutional deep learning model for improving mammographic breast-microcalcification diagnosis[J/OL]. Scientific Reports, 2021, 11(1): 23925. https://doi.org/10.1038/s41598-021-03516-0.
[30]	MUTASA S, CHANG P, NEMER J, et al. Prospective Analysis Using a Novel CNN Algorithm to Distinguish Atypical Ductal Hyperplasia From Ductal Carcinoma in Situ in Breast[J/OL]. Clinical breast cancer, 2020, 20(6): e757-e760. https://doi.org/10.1016/j.clbc.2020.06.001.

BI-RADS	良恶性	例数	百分比
2	肯定为良性	3,180	44.32%
3	良性可能性大，建议随访	1,751	24.40%
4A-C	可疑恶性，建议穿刺活检	1,342	18.70%
5	肯定为恶性	902	12.58%
合计		7,175	100.00%

	病理		百分比（%）	灵敏度	特异度	准确度	AUC（95%CI）
	阳性	阴性	百分比（%）	灵敏度	特异度	准确度	AUC（95%CI）
SDr+JDr
阳性	308	115	68.89	0.885	0.559	0.745	0.73(0.69-0.76)
阴性	40	146	30.29
SDr+JDr+DL
阳性	323	81	65.80	0.926	0.690	0.825	0.81(0.78-0.84)
阴性	26	180	33.55				P=0.014 Z=2.46
JDr+JDr
阳性	301	118	68.24	0.878	0.548	0.735	0.72(0.69-0.76)
阴性	42	143	30.13
JDr+JDr+DL
阳性	310	85	64.33	0.896	0.674	0.801	0.79(0.76-0.82)
阴性	36	176	34.53				P=0.049 Z=1.97

	钙化病灶遗漏数（例）	钙化病灶遗漏率（%）	P值	Z值
DL	7	1.14%
JDr+JDr	10	1.63%
JDr+JDr+DL	7	1.14%	0.46	0.73
SDr+JDr	5	0.81%
SDr+JDr+DL	4	0.65%	0.74	0.34

病理分类	例数	准确率（%）		Z值	P值	准确率（%）		Z值	P值
病理分类	例数	SDr+JDr+DL	SDr+JDr	Z值	P值	JDr+JDr+DL	JDr+JDr	Z值	P值
纤维腺瘤	95	64.21	54.74	1.33	0.18	67.37	49.47	2.50	0.01
腺病	92	67.39	54.35	1.81	0.07	66.30	58.70	1.07	0.29
其他良性肿瘤	45	75.56	53.33	2.20	0.03	64.44	48.89	1.49	0.14
浸润性非特殊类型癌	206	95.15	93.20	0.84	0.40	93.20	92.72	0.19	0.85
导管原位癌	133	84.96	76.69	1.71	0.09	78.20	72.93	1.00	0.32
导管内乳头状瘤	30	80.00	70.00	0.89	0.37	76.67	70.00	0.58	0.56
其他恶性肿瘤	13	100	100	NA	NA	100	92.31	1.02	0.31

基于深度学习乳腺X线摄影钙化识别分类模型的临床应用价值

Assessing the Clinical Utility of a Deep Learning-Based Model for Calcification Recognition and Classification in Mammograms

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 10

参考文献 30

相关文章 15

编辑推荐

Metrics

本文评价

[1]	王子元, 王国中. 改进的轻量级YOLOv5算法在行人检测的应用[J]. 数据与计算发展前沿, 2023, 5(6): 161-172.
[2]	巨家骥, 黄勃, 张帅, 郭茹燕. 融合情感词典和自注意力的双通道情感分析模型[J]. 数据与计算发展前沿, 2023, 5(4): 101-111.
[3]	李俊飞, 徐黎明, 汪洋, 魏鑫. 基于深度学习技术的科技文献引文分类研究综述[J]. 数据与计算发展前沿, 2023, 5(4): 86-100.
[4]	李妍,何洪波,王闰强. 微博热度预测研究综述[J]. 数据与计算发展前沿, 2023, 5(2): 119-135.
[5]	刘云帆,李琦,孙哲南,谭铁牛. 基于生成对抗网络的人脸年龄编辑方法综述[J]. 数据与计算发展前沿, 2023, 5(2): 2-23.
[6]	涂又友,郑奇靖,赵瑾. 基于深度学习方法研究分子/固体界面量子化质子耦合的电荷转移过程[J]. 数据与计算发展前沿, 2023, 5(2): 37-49.
[7]	许淞源,刘峰. ESDRec：一种面向地球大数据平台的数据推荐模型[J]. 数据与计算发展前沿, 2023, 5(1): 55-64.
[8]	陈琼,杨咏,黄天林,冯媛. 小样本图像语义分割综述[J]. 数据与计算发展前沿, 2021, 3(6): 17-34.
[9]	蒲晓蓉,黄佳欣,刘军池,孙家瑜,罗纪翔,赵越,陈柯成,任亚洲. 面向临床需求的CT图像降噪综述[J]. 数据与计算发展前沿, 2021, 3(6): 35-49.
[10]	何涛,王桂芳,马廷灿. 基于词嵌入语义异常的跨学科研究内容发现方法[J]. 数据与计算发展前沿, 2021, 3(6): 50-59.
[11]	张怡宁,何洪波,王闰强. 热门数字音频预测技术综述[J]. 数据与计算发展前沿, 2021, 3(4): 81-92.
[12]	陈子健,李俊,岳兆娟,赵泽方. 基于自编码器与属性信息的混合推荐模型[J]. 数据与计算发展前沿, 2021, 3(3): 148-155.
[13]	肖建平,龙春,赵静,魏金侠,胡安磊,杜冠瑶. 基于深度学习的网络入侵检测研究综述[J]. 数据与计算发展前沿, 2021, 3(3): 59-74.
[14]	李序,连一峰,张海霞,黄克振. 网络安全知识图谱关键技术[J]. 数据与计算发展前沿, 2021, 3(3): 9-18.
[15]	赵伟昱,张宏海,仲波. 基于深度学习的遥感影像地块分割方法[J]. 数据与计算发展前沿, 2021, 3(2): 133-141.