目的: 本研究旨在探讨基于多模态超声特征的卷积神经网络（convolutional neural networks,CNN）在乳腺癌分子亚型诊断中的应用,并评估其在提高诊断效能上的潜力。方法: 通过集成常规超声、应变式弹性成像和剪切波弹性成像三种超声图像的特征,采用不同的超声模态组合分别构建CNN模型,采用5折交叉验证的形式对CNN模型进行训练和验证,采用精确度、召回率、F1分数和曲线下面积（AUC）指标对模型诊断效能进行评估,并进行亚型预测混淆矩阵分析。结果: 共收集病例90例。基于三种超声模态的CNN模型在乳腺癌分子亚型分类的精确度、召回率、F1分数和AUC指标上的表现均较满意（分别为0.914,0.923,0.913,和0.942）,且显著高于基于单一超声模态以及基于任意两种超声模态组合的CNN模型。混淆矩阵分析显示基于三种超声模态的CNN模型对各个亚型的预测准确性最高。结论: 基于多模态超声特征的CNN模型在乳腺癌分子亚型诊断中具有很大潜力,可为乳腺癌的精准诊断和治疗提供有力支持。
Abstract
Objective This study aimed to investigate the application of convolutional neural networks (CNN) based on multimodal ultrasound features in the diagnosis of molecular subtypes of breast cancer, and to assess its potential in improving diagnostic performance. Methods By integrating features from conventional ultrasound, strain elastography, and shear wave elastography, CNN models were built using different combinations of ultrasound modalities. The CNN models were trained and validated through 5-fold cross-validation, and model performance was assessed using accuracy, recall, F1 score, and area under the curve (AUC) metrics, along with subtype prediction confusion matrix analysis. Results A total of 90 cases were collected. The CNN model based on three types of ultrasound modalities achieved satisfactory performance in terms of accuracy, recall, F1 score, and AUC indicators for the classification of molecular subtypes of breast cancer (0.914, 0.923, 0.91, and 0.942, respectively), which was significantly higher than the CNN models based on a single ultrasound modality or any two combined ultrasound modalities. Confusion matrix analysis showed that the CNN model based on three types of ultrasound modalities had the highest predictive accuracy for each subtype. Conclusion The CNN model based on multimodal ultrasound features has great potential in the diagnosis of molecular subtypes of breast cancer and can provide strong support for the precise diagnosis and treatment of breast cancer.
关键词
乳腺癌 /
分子亚型 /
多模态超声特征 /
卷积神经网络
Key words
breast cancer /
molecular subtypes /
multimodal ultrasound features /
convolutional neural networks
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] GIULIANO AE, CONNOLLY JL, EDGE SB, et al.Breast Cancer-Major changes in the American Joint Committee on Cancer eighth edition cancer staging manual[J]. CA Cancer J Clin, 2017, 67(4): 290-303.
[2] UDDIN MN, WANG X.Identification of Breast Cancer Subtypes Based on Gene Expression Profiles in Breast Cancer Stroma[J]. Clin Breast Cancer, 2022, 22(6): 521-537.
[3] 中国抗癌协会乳腺癌专业委员会. 中国抗癌协会乳腺癌诊治指南与规范 (2021年版)[J]. 中国癌症杂志, 2021, 31(10): 954-1040.
[4] NAGARAJ G, MA CX.Clinical Challenges in the Management of Hormone Receptor-Positive, Human Epidermal Growth Factor Receptor 2-Negative Metastatic Breast Cancer: A Literature Review[J]. Adv Ther, 2021, 38(1): 109-136.
[5] 吴毅, 张小勤. 人工智能在医学图像处理中的研究进展与展望[J]. 第三军医大学学报, 2021, 43(18): 1707-1712.
[6] LITJENS G, KOOI T, BEJNORDI BE, et al.A survey on deep learning in medical image analysis[J]. Med Image Anal, 2017, 42: 60-88.
[7] 王瑞琦, 马晓娟, 牛逸凡, 等. 人工智能在乳腺癌超声诊断中的应用进展[J]. 中国医药, 2023, 18(06): 945-948.
[8] WU W, PAN Y.Adaptive Modular Convolutional Neural Network for Image Recognition[J]. Sensors (Basel), 2022, 22(15): 5488-5499.
[9] LECUN Y, BENGIO Y, HINTON G.Deep learning[J]. Nature, 2015, 521: 436-444.
[10] 邱远蓬. 不同分子亚型乳腺癌患者保乳术后放疗生存获益比较及复发危险因素分析[J]. 山西医药杂志, 2022, 51(03): 301-304.
[11] HAN S, KANG HK, JEONG JY, et al.A deep learning framework for supporting the classification of breast lesions in ultrasound images[J]. Phys Med Biol, 2017, 62(19): 7714-7728.
[12] BYRA M, GALPERIN M, OJEDA-FOURNIER H, et al.Breast mass classification in sonography with transfer learning using a deep convolutional neural network and color conversion[J]. Med Phys, 2019, 46(2): 746-755.
[13] SOKOLOVA M, LAPALME G.A systematic analysis of performance measures for classification tasks[J]. Information Processing & Management, 2009, 45(4): 427-437.
[14] 赵巧英, 赵冰, 吴刚, 等. 超声剪切波弹性成像定量参数对乳腺癌系统性治疗后局部区域复发的预测价值[J]. 中华实用诊断与治疗杂志, 2023, 37(4):377-382.
[15] 白连伟, 白莲艳, 邢德君, 等. 老年乳腺癌患者分子分型的临床特点及预后分析[J]. 吉林医学, 2023, 44(03): 667-672.
[16] LITJENS G, KOOI T, BEJNORDI BE, et al.A survey on deep learning in medical image analysis[J]. Med Image Anal, 2017, 42: 60-88.
基金
湖南省卫生健康委2023年度科研计划课题“多模态超声特征与乳腺癌不同分子亚型的相关性研究”（D202309027332）
PDF(2367 KB)
