人工智能在乳腺癌影像诊疗中的研究进展
2024-10-30翟天旭张敏伟李德春
摘要:乳腺癌是全球最常见的癌症之一,也是导致女性癌症死亡的主要原因。近年来,随着计算机性能的飞速提升,人工智能在各个领域中大放异彩,拥有自动分析图象能力的人工智能深度学习在医疗领域同样引起越来越多的关注,医疗机构开始重视医疗数据的收集,尤其是大量的医学影像资料的积累。当前对于乳腺疾病的常规影像学检查方法主要有3种:乳腺X线摄影、乳腺超声检查和乳腺MRI检查。人工智能结合乳腺影像学检查为乳腺癌的诊疗提供了前所未有的机会。本文现从人工智能和乳腺影像数据相结合在乳腺癌的诊断、治疗及预后预测等方面进行综述,以期将人工智能更加广泛、成熟地运用于乳腺癌的影像诊疗中,为推动人工智能结合乳腺影像数据实现乳腺癌精准医疗由理论到临床实践的转化应用提供思路。
关键词:人工智能;乳腺癌;乳腺X线摄影;超声;磁共振
Research progress of artificial intelligence in imaging diagnosis and treatment of breast cancer
ZHAI Tianxu1, ZHANG Minwei2, LI Dechun2
1Department of Radiology, Nanyang First People's Hospital, Nanyang 473000, China; 2Department of Radiology, Xuzhou Central Hospital, Xuzhou 221009, China
Abstract: Breast cancer is one of the most common cancers worldwide and the leading cause of cancer death in women.In recent years, with the rapid improvement of computer performance, artificial intelligence shines in various fields, and artificial intelligence deep learning with automatic image analysis ability has also attracted more and more attention in the medical field, medical institutions have begun to pay attention to the collection of medical data, especially the accumulation of a large number of medical image data. At present, there are three conventional imaging methods for breast diseases: mammography, breast ultrasound and breast MRI.Artificial intelligence combined with breast imaging offers unprecedented opportunities for the diagnosis and treatment of breast cancer.This paper reviews the combination of artificial intelligence and breast image data in the diagnosis, treatment and prognosis prediction of breast cancer, hoping that artificial intelligence can be more widely and maturely applied to the imaging diagnosis and treatment of breast cancer, so as to provide ideas for promoting the transformation and application of precision medicine for breast cancer from theory to clinical practice by combining artificial intelligence with breast image data.
Keywords: artificial intelligence; breast cancer; mammography; ultrasound; magnetic resonance
乳腺癌精准医疗的实现离不开人工智能的参与[1],基于人工智能的影像组学分析已经在乳腺癌的诊断、治疗以及预后评估方面取得了一定的成绩。乳腺癌是导致女性癌症死亡的主要原因[2, 3],早期发现和治疗对于提高患者的生存率至关重要[4]。在两癌筛查计划稳步推进的大背景下,乳腺癌筛查人数的大量增加、乳腺癌发病率的逐年提高以及专业乳腺钼靶诊断医生的短缺已成为目前放射科和待筛查人群共同面临的严峻挑战[5, 6]。人工智能助力影像诊疗是解决这一问题的首选方法[7],同时人工智能技术可以通过分析患者的临床数据和影像学检查结果,为每位患者提供个性化的治疗方案[8]。
当前对于乳腺疾病的常规影像学检查方法主要有3种:乳腺X线摄影、乳腺超声检查和乳腺MRI检查[9]。根据疾病类型和具体情况,在3种检查方法中选择最适合的影像学检查方法至关重要,因为不同的成像技术各有优势和限制,正确的选择不仅有助于节约资源,还可以提高诊断准确性[10]。人工智能在3种检查方法中的应用各有侧重,本文将系统性地探讨人工智能联合乳腺影像学检查在乳腺癌诊断、治疗和管理中的应用现状及未来发展。
1" 人工智能联合乳腺X线诊断乳腺癌
乳线X线摄影检查是用于乳腺癌早期发现的使用最广泛的工具之一[11],乳腺X线摄影具有价格便宜、操作简单、处理速度快等优点,是全球女性乳腺癌筛查中最常用的检查方法[12]。有研究表明乳腺X线摄影检查是评估乳腺钙化状态最可靠的技术,可将乳腺癌死亡率降低高达30%[13, 14]。但在当前的临床实践中,放射科医师对乳腺X光摄影图像的主观解释各不相同,只有59.0%和63.0%的放射科医师在乳腺X线摄影筛查中达到了推荐的异常判读率和特异性水平[11, 15]。此外双重阅读(即由高年资放射科医生审核低年资医生已经初步诊断过的乳腺X光片)也用以提高放射科医生的表现,虽然双重阅读使放射科医生在诊断敏感性方面有所改善,但也降低了工作效率,并增加了假阳性检查结果[13, 16]。在目前专业放射科医师短缺的情况下,由于放射科医师的工作量较大,放射科医师的表现可能会在更大规模的乳腺癌X线筛查中受到进一步影响[15, 17]。人工智能联合乳腺X线摄影主要用于提高乳腺癌检出的准确度。有学者通过构建基于乳腺X线摄影的放射组学列线图,发现深度学习模型对乳腺钙化良恶性的诊断准确率为0.835,与高年资放射科医师的诊断效能相当,是区分良性钙化和恶性钙化的潜在工具[18]。有研究同样证实了人工智能可以较显著的提高乳腺癌大规模筛查中乳腺癌检出的准确率,而且还将低年资放射科医师在接受人工智能辅助前后的表现进行对比,发现在人工智能的辅助下,低年资放射科医师的诊断性能有了明显的提高[19]。在一项亚洲女性乳腺密度与种族差异相关性的横断面研究中发现,中国女性的乳腺密度较其他亚洲国家女性(如马来西亚、韩国等)的乳腺密度更高[20, 21]。乳腺癌的发病率随着乳腺密度的增高明显上升,放射科医师筛查乳腺癌的敏感度会随着乳腺密度的增加而降低[22, 23],有研究表明人工智能在乳腺密度较高的女性中,检出乳腺癌的准确性明显高于放射科医生[24]。这些研究成果充分肯定了人工智能技术联合乳腺X线摄影在乳腺癌诊断中的作用和可靠性。因此,人工智能联合乳腺X线摄影可为乳腺癌的诊断提供可靠的依据。
2" 人工智能联合乳腺超声诊断乳腺癌
乳腺超声由于无放射性,可适用于任何年龄,特别是妊娠及哺乳期女性的乳腺检查,是一种具有无创实用、可重复性强的乳腺检查方法[25, 26]。有研究通过内部测试数据集和外部验证数据集中早期乳腺癌与良性病变在超声中的表现进行评估,发现放射科医生在人工智能辅助下的诊断效率显着高于没有人工智能协助的放射科医生[26-28]。EDL-BC深度学习模型可以识别乳腺病变超声图像上细微但信息丰富的元素,从而显着提高放射科医生识别早期乳腺癌的诊断性能,使患者在临床实践中受益。有学者使用智能多模式剪切波弹性成像(SWE)探讨是否可以在不影响乳腺癌检出率的情况下减少不必要的活检数量,发现智能多模态 SWE 的敏感度为100%,特异度为50.3%,曲线下面积为0.93,与传统的SWE和B型乳腺超声相比,诊断性能明显更高[29-31]。智能多模式SWE与B型超声相比,不必要的活检减少了50.3%,且未遗漏癌症,智能多模式SWE可以安全地避免大多数不必要的乳腺活检,这可能有助于减轻患者和医疗保健系统的负担[32]。有学者认为乳腺癌的超声诊断通常基于来自单一超声模式的整个乳腺肿瘤的单个区域,这限制了诊断性能,而乳腺肿瘤的多模态超声图像上的多个区域都可能包含有用的诊断信息,从而提出了一种多区域放射组学方法结合多模式超声,用于恶性和良性乳腺肿瘤的人工智能诊断,发现除整个肿瘤区域外,超声图像上灌注最强的区域、边缘区域和周围区域均能辅助诊断乳腺癌,证实了人工智能联合乳腺超声能提高乳腺癌检出的敏感率[33-36]。
3" 人工智能联合乳腺MRI诊断乳腺癌
乳腺MRI检查的优势在于无辐射性且有较高的软组织分辨率,可以较敏感的发现乳腺病变[33, 37, 38]。有学者开发了一个卷积神经网络以增强MRI上乳腺病变的计算机化检测,发现混合3D/2D U-Net 架构在乳腺肿瘤分割方面的取得了优异的表现,证明了人工智能在准确识别乳腺MRI图像上是否存在病变的可行性[39]。有研究同样认为基于MRI图像的深度学习分析有可能提高乳腺癌的诊断准确性,动态增强检查还可以根据病变血流灌注情况帮助临床医生对良恶性疾病进行鉴别,最终为患者带来更好的治疗效果[40-42]。有学者使用高时间分辨率-动态对比增强(HTR-DCE)MR序列评估放射组学分析的诊断性能,以区分良性和恶性乳腺病变,通过提取半定量增强参数和纹理特征,计算HTR-DCE MR序列中每个阶段的纹理特征的时间变化(“动力学纹理参数”),使用LASSO逻辑回归和交叉验证进行统计分析以建立模型,证实使用HTR-DCE的放射组学分析比传统乳腺MRI具有更好的诊断性能(AUC=0.876),从HTR-DCE-MRI中提取的动力学纹理特征在区分良性和恶性乳腺病变方面发挥着重要作用[43-45]。因此人工智能和多模态磁共振以及动态增强磁共振的联合使用将为乳腺疾病的评估提供更加可靠的依据。
4" 人工智能助力乳腺癌的治疗和预后评估
乳腺癌的发生发展与基因组学密切相关,某些基因突变与乳腺癌的易感性和预后密切相关[46]。基于深度学习的基因模型可以预测乳腺癌的发生风险,并指导患者进行个性化的预防和治疗[47, 48]。有学者通过对cfDNA的甲基化分析,将全局甲基化模式深度学习算法和基于标记的算法相结合来检测癌症,发现基于选定标记和全局甲基化模式的深度学习模型的测试数据准确性分别为0.88和0.90,AUC为0.90和0.96,认为在早期癌症检测和乳腺癌治疗领域,深度学习模型具有巨大的潜力[47]。有研究通过使用基于人工智能的纵向多参数MRI来预测乳腺癌新辅助化疗后能否达到完全病理缓解,提取患者新辅助化疗前后所作MRI的放射组学特征,开发了5个机器学习分类器来准确预测每个亚型乳腺癌能否完全病理缓解,建立了一种新的有卓越性能的工具来预测乳腺癌对新辅助化疗的反应, 有助于确定乳腺癌术后新辅助化疗手术策略[49, 50]。有学者通过一项大规模的多中心研究,证实人工智能技术可以帮助医生对乳腺癌患者的及预后评估提供可靠的指导[51]。
5" 小结与展望
人工智能技术在乳腺癌的诊断、治疗和预后评估中具有巨大的潜力,可以提高诊断准确性、个性化治疗效果和医疗资源利用效率,为乳腺癌患者带来更好的医疗服务和生活质量。然而,人工智能技术的应用还面临着一些挑战,包括数据隐私保护、算法可解释性、临床实用性等方面的问题,未来仍需不断加强技术研究和临床实践,促进人工智能技术在乳腺癌诊疗中的广泛应用和持续发展。
在人工智能飞速发展的背景下,可以期待以下方面的发展:第一,多模态数据融合,将临床数据、影像学检查结果、基因组学数据等多模态数据进行融合分析,提高乳腺疾病诊断和治疗的准确性和效率。第二,深度学习模型优化,不断优化深度学习模型的结构和算法,提高模型在乳腺疾病诊断、治疗和管理中的性能,减少误诊和漏诊的风险。第三,进一步推进个性化医疗模式,根据患者的个体特征和病情变化,提供定制化的诊疗方案,最大程度地提高治疗效果和生存质量。第四,通过人工智能技术的应用,优化医疗资源的分配和利用,提高医疗服务的效率和质量,降低医疗成本,使更多的患者能够享受到优质的医疗服务。第五,加强国际合作与交流,共同推动人工智能技术在乳腺疾病领域的应用和发展,分享经验和技术成果,促进全球乳腺健康事业的发展和进步。通过持续不断地努力和合作,期待在未来实现乳腺疾病诊断、治疗和管理的精准化、个性化和智能化,为患者带来更好的医疗服务和生活质量。
参考文献:
[1]" " 中国科学院上海营养与健康研究所/中国科学院上海生命科学信息中心团队, 复旦大学团队. 趋势观察: 精准医学领域发展态势分析[J]. 中国科学院院刊, 2023, 38(6): 935-42.
[2]" "Harbeck N, Gnant M. Breast cancer[J]. Lancet, 2017, 389(10074): 1134-50.
[3]" "Sung H, Ferlay J, Siegel RL, et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries[J]. CA Cancer J Clin, 2021, 71(3): 209-49.
[4]" "Trayes KP, Cokenakes SEH. Breast cancer treatment[J]. Am Fam Physician, 2021, 104(2): 171-8.
[5]" "Winters S, Martin C, Murphy D, et al. Breast cancer epidemiology, prevention, and screening[J]. Prog Mol Biol Transl Sci, 2017, 151: 1-32.
[6]" "Yamashita R, Long J, Longacre T, et al. Deep learning model for the prediction of microsatellite instability in colorectal cancer: a diagnostic study[J]. Lancet Oncol, 2021, 22(1): 132-41.
[7]" "Yala A, Schuster T, Miles R, et al. A deep learning model to triage screening mammograms: a simulation study[J]. Radiology, 2019, 293(1): 38-46.
[8]" "Zhu X, Wolfgruber TK, Leong L, et al. Deep learning predicts interval and screening‑detected cancer from screening mammograms: a case‑case‑control study in 6369 women[J]. Radiology, 2021, 301(3): 550-8.
[9]" "周雅秋, 张淑平, 刘雅聪, 等. 不同乳腺影像学检查及其新技术对乳腺癌新辅助治疗评估的价值与比较[J]. 临床放射学杂志, 2024, 43(1): 153-6, 封3.
[10]" Andersson I, Janzon L, Sigfússon BF. Mammographic breast cancer screening: a randomized trial in Malmö, Sweden[J]. Maturitas, 1985, 7(1): 21-9.
[11]" Tabár L, Yen AM, Wu WY, et al. Insights from the breast cancer screening trials: how screening affects the natural history of breast cancer and implications for evaluating service screening programs[J]. Breast J, 2015, 21(1): 13-20.
[12] Hong ZL, Chen S, Peng XR, et al. Nomograms for prediction of breast cancer in breast imaging reporting and data system (BI-RADS) ultrasound category 4 or 5 lesions: a single‑center retrospective study based on radiomics features[J]. Front Oncol, 2022, 12: 894476.
[13] Liu HH, Chen YH, Zhang YZ, 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]. Eur Radiol, 2021, 31(8): 5902-12.
[14] Lotter W, Diab AR, Haslam B, et al. Robust breast cancer detection in mammography and digital breast tomosynthesis using an annotation-efficient deep learning approach[J]. Nat Med, 2021, 27(2): 244-9.
[15] Lei CQ, Wei W, Liu ZY, et al. Mammography-based radiomic analysis for predicting benign BI-RADS category 4 calcifications[J]. Eur J Radiol, 2019, 121: 108711.
[16] Wang GS, Shi DF, Guo Q, et al. Radiomics based on digital mammography helps to identify mammographic masses suspicious for cancer[J]. Front Oncol, 2022, 12: 843436.
[17] Baccouche A, Garcia-Zapirain B, Zheng YF, et al. Early detection and classification of abnormality in prior mammograms using image‑to‑image translation and YOLO techniques[J]. Comput Methods Programs Biomed, 2022, 221: 106884.
[18] Al-Masni MA, Al-Antari MA, Park JM, et al. Simultaneous detection and classification of breast masses in digital mammograms via a deep learning YOLO-based CAD system[J]. Comput Methods Programs Biomed, 2018, 157: 85-94.
[19] Salman ME, Çakirsoy Çakar G, Azimjonov J, et al. Automated prostate cancer grading and diagnosis system using deep learning-based Yolo object detection algorithm[J]. Expert Syst Appl, 2022, 201: 117148.
[20] Mariapun S, Li JM, Yip CH, et al. Ethnic differences in mammographic densities: an Asian cross-sectional study[J]. PLoS One, 2015, 10(2): e0117568.
[21]" Ji Y, Li BX, Zhao R, et al. The relationship between breast density, age, and mammographic lesion type among Chinese breast cancer patients from a large clinical dataset[J]. BMC Med Imaging, 2021, 21(1): 43.
[22] Nishiyama K, Taira N, Mizoo T, et al. Influence of breast density on breast cancer risk: a case control study in Japanese women[J]. Breast Cancer, 2020, 27(2): 277-83.
[23]" 何思怡, 李" 贺, 曹毛毛, 等. 全球及我国女性乳腺癌疾病负担年龄分布及变化趋势[J]. 中国肿瘤, 2023, 32(1): 1-7.
[24] Park GE, Kim SH, Lee JM, et al. Comparison of positive predictive values of categorization of suspicious calcifications using the 4th and 5th editions of BI-RADS[J]. AJR Am J Roentgenol, 2019, 213(3): 710-5.
[25] 靳凤梅, 王" 惠, 马苏美, 等. 乳腺癌ABVS表现与ER、PR、Her-2及Ki67的相关性研究[J]. 中国超声医学杂志, 2018, 34(2): 115-9.
[26] Gao Y, Zeng SQ, Xu XY, et al. Deep learning‑enabled pelvic ultrasound images for accurate diagnosis of ovarian cancer in China: a retrospective, multicentre, diagnostic study[J]. Lancet Digit Health, 2022, 4(3): e179-e187.
[27] Vachon CM, Scott CG, Norman AD, et al. Impact of artificial intelligence system and volumetric density on risk prediction of interval, screen‑detected, and advanced breast cancer[J]. J Clin Oncol, 2023, 41(17): 3172-83.
[28]" Zhang JD, Wu JJ, Zhou XS, et al. Recent advancements in artificial intelligence for breast cancer: image augmentation, segmentation, diagnosis, and prognosis approaches[J]. Semin Cancer Biol, 2023, 96: 11-25.
[29] Liao JW, Gui Y, Li ZL, et al. Artificial intelligence-assisted ultrasound image analysis to discriminate early breast cancer in Chinese population: a retrospective, multicentre, cohort study[J]. EClinicalMedicine, 2023, 60: 102001.
[30] Bartolotta TV, Orlando AAM, Dimarco M, et al. Diagnostic performance of 2D-shear wave elastography in the diagnosis of breast cancer: a clinical appraisal of cutoff values[J]. Radiol Med, 2022, 127(11): 1209-20.
[31]" Zhou JQ, Zhan WW, Chang C, et al. Breast lesions: evaluation with shear wave elastography, with special emphasis on the \"stiff rim\" sign[J]. Radiology, 2014, 272(1): 63-72.
[32] Chen YL, Chang C, Zeng W, et al. 3-Dimensional shear wave elastography of breast lesions: added value of color patterns with emphasis on crater sign of coronal plane[J]. Medicine, 2016, 95(39): e4877.
[33] Xu Z, Wang YQ, Chen M, et al. Multi-region radiomics for artificially intelligent diagnosis of breast cancer using multimodal ultrasound[J]. Comput Biol Med, 2022, 149: 105920.
[34] Berg WA, López Aldrete AL, Jairaj A, et al. Toward AI-supported US triage of women with palpable breast lumps in a low-resource setting[J]. Radiology, 2023, 307(4): e223351.
[35] Zhou BY, Wang LF, Yin HH, et al. Decoding the molecular subtypes of breast cancer seen on multimodal ultrasound images using an assembled convolutional neural network model: a prospective and multicentre study[J]. EBioMedicine, 2021, 74: 103684.
[36] Lee SH, Cho N, Chang JM, et al. Two-view versus single-view shear-wave elastography: comparison of observer performance in differentiating benign from malignant breast masses[J]. Radiology, 2014, 270(2): 344-53.
[37] Bakker MF, de Lange SV, Pijnappel RM, et al. Supplemental MRI screening for women with extremely dense breast tissue[J]. N Engl J Med, 2019, 381(22): 2091-102.
[38] Saslow D, Boetes C, Burke W, et al. American Cancer Society guidelines for breast screening with MRI as an adjunct to mammography[J]. CA Cancer J Clin, 2007, 57(2): 75-89.
[39]" Ma XY, Wang JL, Zheng XP, et al. Automated fibroglandular tissue segmentation in breast MRI using generative adversarial networks[J]. Phys Med Biol, 2020, 65(10): 105006.
[40] Kim JJ, Kim JY, Suh HB, et al. Characterization of breast cancer subtypes based on quantitative assessment of intratumoral heterogeneity using dynamic contrast‑enhanced and diffusion-weighted magnetic resonance imaging[J]. Eur Radiol, 2022, 32(2): 822-33.
[41]Adam R, Dell'Aquila K, Hodges L, et al. Deep learning applications to breast cancer detection by magnetic resonance imaging: a literature review[J]. Breast Cancer Res, 2023, 25(1): 87.
[42]" 许" 华, 陈士新, 付" 伟, 等. 磁共振动态对比增强及弥散加权成像预测乳腺癌前哨淋巴结转移的价值[J]. 分子影像学杂志, 2020, 43(4): 721-5.
[43] Geuzinge HA, Heijnsdijk EAM, Obdeijn IM, et al. Experiences, expectations and preferences regarding MRI and mammography as breast cancer screening tools in women at familial risk[J]. Breast, 2021, 56: 1-6.
[44] Comstock CE, Gatsonis C, Newstead GM, et al. Comparison of abbreviated breast MRI vs digital breast tomosynthesis for breast cancer detection among women with dense breasts undergoing screening[J]. JAMA, 2020, 323(8): 746-56.
[45] Perre SV, Duron L, Milon A, et al. Radiomic analysis of HTR-DCE MR sequences improves diagnostic performance compared to BI-RADS analysis of breast MR lesions[J]. Eur Radiol, 2021, 31(7): 4848-59.
[46] 中国医师协会精准治疗委员会乳腺癌专业委员会, 中华医学会肿瘤学分会乳腺肿瘤学组, 中国抗癌协会乳腺癌专业委员会等. 中国乳腺癌患者BRCA1/2基因检测与临床应用专家共识(2018年版). 中国癌症杂志. 2018, 28(10): 787-800.
[47] Bahado-Singh RO, Turkoglu O, Aydas B, et al. Precision oncology: artificial intelligence, circulating cell-free DNA, and the minimally invasive detection of pancreatic cancer‑a pilot study[J]. Cancer Med, 2023, 12(19): 19644-55.
[48] Dogan H, Hakguder Z, Madadjim R, et al. Elucidation of dynamic microRNA regulations in cancer progression using integrative machine learning[J]. Brief Bioinform, 2021, 22(6): bbab270.
[49] Huang YH, Zhu T, Zhang XL, et al. Longitudinal MRI-based fusion novel model predicts pathological complete response in breast cancer treated with neoadjuvant chemotherapy: a multicenter, retrospective study[J]. EClinicalMedicine, 2023, 58: 101899.
[50] Thompson BM, Chala LF, Shimizu C, et al. Pre-treatment MRI tumor features and post-treatment mammographic findings: may they contribute to refining the prediction of pathologic complete response in post-neoadjuvant breast cancer patients with radiologic complete response on MRI?[J]. Eur Radiol, 2022, 32(3): 1663-75.
[51] Qi TH, Hian OH, Kumaran AM, et al. Multi-center evaluation of artificial intelligent imaging and clinical models for predicting neoadjuvant chemotherapy response in breast cancer[J]. Breast Cancer Res Treat, 2022, 193(1): 121-38.
(编辑:林" 萍)
收稿日期:2024-04-25
基金项目:江苏省十四五医学重点学科项目(ZDXK202237);徐州市科学技术局社会发展项目(KC15SH061)
作者简介:翟天旭,硕士,住院医师,E-mail: 382852742@qq.com
通信作者:李德春,硕士,主任医师,E-mail: 18952171358@189.cn