Value of ultrasound-based radiomics in identifying benign and malignant BI-RADS category 4a irregular breast lesions and reducing unnecessary biopsies
-
摘要:
目的 探讨超声影像组学对BI-RADS 4a类不规则乳腺结节良恶性的鉴别价值，并结合影像组学、超声特征及临床独立危险因素特征建立列线图，评估其在减少不必要活检中的价值。 方法 回顾性收集常规超声检查筛选出的BI-RADS 4a类不规则乳腺结节905例，随机分为训练队列（n=634）和验证队列（n=271），比例为7∶3。共收集851个影像组学特征，以手术病理结果为金标准，通过Logistic回归模型构建影像组学模型，同时利用单因素逻辑分析及多因素逻辑分析结合影像组学特征、超声特征及临床独立危险因素建立影像组学模型，通过ROC曲线评估影像组学模型及列线图模型对超声BI-RADS 4a类形态不规则乳腺结节的诊断效能。 结果 905例不规则乳腺结节中，恶性结节485个，良性结节420个；患者年龄22~83（50.05±11.13）岁，训练队列及验证队列的年龄、Rad-score值、肿块直径等结果差异无统计学意义（P>0.05）；训练队列影像组学模型AUC值为0.927（95% CI：0.900~0.950），验证队列影像组学模型AUC值为0.946（95% CI：0.908~0.976），该模型训练队列的敏感度、准确度、特异性、F1值、精确度分别为0.879、0.879、0.877、0.909、0.940，该模型验证队列的敏感度、准确度、特异性、F1值、精确度分别为0.890、0.896、0.909、0.921、0.956；校准曲线显示该模型训练队列和验证队列有较好的校准度；训练队列列线图模型AUC值为0.943（95% CI：0.912~0.960），验证队列列线图模型AUC值为0.968（95% CI：0.924~0.970）。 结论 超声影像组学及列线图模型在提高BI-RADS 4a类形态不规则乳腺结节良恶性的诊断效能有重要价值，对BI-RADS 4a类不规则乳腺结节有更好的预测效能，并且能够减少不必要的活检。 Abstract:Objective To investigate the value of ultrasound-based radiomics in discriminating benign and malignant BI-RADS category 4a irregular breast nodules, and to assess its value in reducing unnecessary biopsies by establishing a nomogram through combine radiomics, ultrasound features and clinical independent risk factor characteristics. Methods A total of 905 cases of BI- RADS 4a irregular breast nodules screened by conventional ultrasonography were retrospectively collected and randomly divided into a training set (n=634) and a validation set (n=271) with a ratio of 7:3. Pyradiomics was used to extract 851 features.The radiomics model was constructed by Logistics regression model using surgical pathology as the gold standard. The diagnostic efficacy of the radiomics model and the nomogram model on the diagnosis of irregular breast nodules with ultrasound BI-RADS 4a pattern was evaluated by ROC curve. Results Among 905 irregular breast nodules, 485 were malignant nodules and 420 were benign nodules, with the age of 22-83(50.05±11.13) years old. The age, Rad-score value, and mass diameter of the two groups were not significantly different (P>0.05). The AUC value of the training cohort radiomics model was 0.927 (95% CI: 0.900-0.950), and the AUC value of the validation cohort radiomics model was 0.946 (95% CI: 0.908-0.976). The sensitivity, accuracy, specificity, F1 value, and precision for the training cohort of this model were 0.879, 0.879, 0.877, 0.909 and 0.940, respectively. The sensitivity, accuracy, specificity, F1 value, and precision for the validation cohort of this model were 0.890, 0.896, 0.909, 0.921 and 0.956, respectively. The calibration curves showed good calibration between the training and validation cohorts of this mode; The AUC value was 0.943 (95% CI: 0.912-0.960) for the training cohort normogram model and 0.968 (95% CI: 0.924-0.970) for the validation cohort normogram model. Conclusion Ultrasound-based radiomics and the nomogram model have important value in improving the diagnostic efficacy of benign and malignant BIRADS 4a morphologically irregular breast nodules, have better predictive efficacy for BI-RADS 4a irregular breast nodules, and can reduce unnecessary biopsies. -
Key words:
- ultrasonography /
- radiomics /
- breast tumor /
- forecasting /
- biopsy
-
图 1 恶性乳腺肿块超声图像感兴趣区域分割
Figure 1. Segmentation of regions of interest in ultrasound images of breast cancer.
图 3 训练队列及验证队列影像组学模型的预测模型的ROC曲线
Figure 3. ROC curve of prediction model of imaging radiomics model in training cohort (A) and validation cohort (B).
图 4 训练队列（A）及验证队列（B）影像组学预测模型的校准曲线
注: Y轴代表乳腺癌实际情况, X轴代表影像组学模型对恶性乳腺结节风险预测, 对角线的灰色直线是参考线, 即预测值等于实际值, 如果观察值等于预测值则代表预测结果与实际结果一致, 如果预测值大于观察值, 即高估了风险; 如果预测值小于观察值, 即低估了风险; 从构建的两个模型的预测结果看, 两个模型均具有较好的预测能力.
Figure 4. Calibration curve of radiomics prediction model in training cohort (A) and validation cohort (B).
图 5 基于影像组学的Rad-score的构建
A: 19个系数非零的影像组学特征通过LASSO算法和10倍交叉验证进行参数调整得出系数收敛图, 横坐标为Log（λ）, 纵坐标为各个特征在模型中的各自系数，并且全部这些特征具有的系数均不为0; B: 选择19个系数非零的影像组学特征, 通过LASSO回归算法中调整参数λ的选择. 绘制二项偏差与参数log（λ）之间的关系曲线，通过10倍交叉验证, 选择log(λ)=-4.996所对应的λ值，此时最终筛选的最优影像特征的数目将达到最为适宜的状态; C: 训练队列和验证队列不规则乳腺良恶性结节的Rad-score箱图（蓝色代表良性，黄色代表恶性）; D~E: 训练队列和验证队列每个不规则乳腺结节的Rad-score分布（0代表良性, 1代表恶性）.
Figure 5. Construction of Rad-score based on radiomics.
图 7 训练队列及验证队列列线图模型预测模型的ROC曲线
Figure 7. ROC curve of nomograph model prediction model in training group and verification group.
表 1 训练队列和验证队列临床资料
Table 1. Clinical data of training cohort and validation cohort
临床资料 训练队列（n=634） 验证队列（n=271） P 肿块最大直径（cm, Mean±SD） 2.73±1.166 2.883±1.311 0.175 年龄（岁, Mean±SD） 51.884±10.905 50.237±11.283 0.117 Rad-score（Mean±SD） 0±2.302 -0.297±2.965 0.195 病理类型（n） 0.201 恶性 331 154 良性 303 117 
下载: 导出CSV
表 2 基于训练组单变量和多变量逻辑回归分析结果
Table 2. Results of univariate and multivariate logistic regression analysis based on training group
变量 单变量逻辑分析 多变量逻辑分析 OR（95% CI） P OR（95% CI） P 肥胖 1.077(0.488, 3.503) 0.237 横线 横线 口服避孕药 0.947(0.399, 3.180) 0.502 横线 横线 乳腺癌家族史 1.231(0.441, 3.493) 0.009 横线 0.412 吸烟史 1.641 (0.346, 7.491) 0.342 横线 横线 饮酒史 3.291 (0.581, 16.772) 0.183 横线 横线 病变位置 0.735 (0.415, 1.333) 0.284 横线 横线 年龄 1.077 (1.062, 1.121) 0.013 1.071(0.958, 1.159) 0.002 肿块直径 2.431（1.644, 3.527） 0.011 2.778(1.521, 5.517) 0.002 初潮 < 13岁 0.255 (0.133, 0.476) 0.003 1.017 (0.139, 7.161) 0.461 Rad-score 0.272（0.187，0.339) 0.011 0.211(0.114, 0.425) 0.001
下载: 导出CSV
表 3 19个系数非零的影像组学特征
Table 3. Radiomics characteristics of 19 non-zero coefficients
类别 特征 系数 直方图特征（n=1） original_firstorder_Minimum -0.205 纹理特征（n=2） original_glrlm_LongRunHighGrayLevelEmphasis 0.157 original_glrlm_ShortRunLowGrayLevelEmphasis -0.245 小波特征（n=16） wavelet.LH_firstorder_Mean -1.665 wavelet.HH_gldm_DependenceNonUniformityNormalized -0.954 wavelet.HH_glszm_SizeZoneNonUniformityNormalized -0.742 wavelet.LL_glcm_ClusterShade -0.201 wavelet.LL_glrlm_ShortRunLowGrayLevelEmphasis -0.19 wavelet.HH_firstorder_Skewness -0.01 wavelet.HL_glszm_LargeAreaLowGrayLevelEmphasis 0.02 wavelet.LH_glszm_SmallAreaLowGrayLevelEmphasis 0.062 wavelet.LL_glszm_LargeAreaLowGrayLevelEmphasis 0.175 wavelet.LH_gldm_DependenceNonUniformityNormalized 0.249 wavelet.LH_glszm_SizeZoneNonUniformity 0.268 wavelet.HL_firstorder_Mean 0.279 wavelet.HH_glcm_MCC 0.433 wavelet.HL_firstorder_Median 0.436 wavelet.HH_firstorder_Median 0.556 wavelet.HH_glcm_Imc1 0.734
下载: 导出CSV
表 4 超声影像组学模型的诊断效能
Table 4. Diagnostic efficacy of ultrasound imaging radiomics model
评价指标 AUC 准确度 敏感度 特异性 精确度 F1 训练队列 0.927 0.879 0.879 0.877 0.94 0.909 验证队列 0.946 0.896 0.89 0.909 0.956 0.921
下载: 导出CSV
-
[1] 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. doi: 10.3322/caac.21660 [2] 陈万青, 郑荣寿. 中国女性乳腺癌发病死亡和生存状况[J]. 中国肿瘤临床, 2015, 42(13): 668-74. https://www.cnki.com.cn/Article/CJFDTOTAL-ZGZL201513010.htm [3] D'Orsi CJ, Sickles EA, Mendelson EB, et al. ACR BI-RADS atlas, breast imaging reporting and data system. 5th ed. Reston: American College of Radiology, 2013. [4] 孙杨, 林子梅, 骆洁丽, 等. 超声BI-RADS分类对特殊类型乳腺癌的诊断价值[J]. 中华超声影像学杂志, 2022(1): 37-42. [5] 江庆, 郭建锋. 超声影像组学在乳腺癌中的研究进展[J]. 中华医学超声杂志: 电子版, 2020, 17(11): 1145-8. doi: 10.3877/cma.j.issn.1672-6448.2020.11.017 [6] 罗焱文, 朱庆莉. 乳腺癌影像组学研究进展[J]. 协和医学杂志, 2021, 12(6): 983-8. https://www.cnki.com.cn/Article/CJFDTOTAL-XHYX202106020.htm [7] Luo WQ, Huang QX, Huang XW, et al. Predicting breast cancer in breast imaging reporting and data system (BI-RADS) ultrasound category 4 or 5 lesions: a nomogram combining radiomics and BIRADS[J]. Sci Rep, 2019, 9(1): 11921. doi: 10.1038/s41598-019-48488-4 [8] 周嘉音, 尤超, 顾雅佳. 影像组学在乳腺癌的应用研究进展[J]. 国际医学放射学杂志, 2022, 45(2): 174-9. https://www.cnki.com.cn/Article/CJFDTOTAL-GWLC202202010.htm [9] Li Q, et al. Meta-analysis: contrast-enhanced ultrasound versus conventional ultrasound for differentiation of benign and malignant breast lesions[J]. Ultrasound Med Biol, 2018, 44(5): 919-29. doi: 10.1016/j.ultrasmedbio.2018.01.022 [10] 马骥, 张艳, 陈青青, 等. 常规超声联合ABVS在乳腺BI-RADS 4类小结节诊断中的价值[J]. 同济大学学报: 医学版, 2021, 42(6): 825-30. https://www.cnki.com.cn/Article/CJFDTOTAL-TJIY202106016.htm [11] 马梦伟, 秦耿耿, 陈卫国. 影像组学在乳腺癌分子分型中的应用现状及研究进展[J]. 国际医学放射学杂志, 2020, 43(4): 452-6. https://www.cnki.com.cn/Article/CJFDTOTAL-GWLC202004016.htm [12] 李佳伟, 方舟, 周瑾, 等. 浸润性三阴性乳腺癌超声影像组学特征与肿瘤生物学特性的关系研究[J]. 中华超声影像学杂志, 2019(2): 137-43. [13] Lee SE, Han K, Kwak JY, et al. Radiomics of US texture features in differential diagnosis between triple-negative breast cancer and fibroadenoma[J]. Sci Rep, 2018, 8(1): 13546. [14] 王瑛, 陈英格, 叶素敏, 等. 超声影像组学标签预测三阴性乳腺癌的价值[J]. 中国医学影像学杂志, 2021, 29(1): 35-8, 46. https://www.cnki.com.cn/Article/CJFDTOTAL-ZYYZ202101012.htm [15] Yu FH, et al. Ultrasound-based radiomics nomogram: a potential biomarker to predict axillary lymph node metastasis in early-stage invasive breast cancer[J]. Eur J Radiol, 2019, 119: 108658. [16] Chen X, Huang Y, He L, et al. CT-based radiomics to differentiate pelvic rhabdomyosarcoma from yolk sac tumors in children[J]. Front Oncol, 2020, 10: 584272. [17] Zhang L, Jiang XH, Xie XM, et al. The impact of preoperative radiomics signature on the survival of breast cancer patients with residual tumors after NAC[J]. Front Oncol, 2021, 10: 523327. [18] Xie JK, Li BS, Min XD, et al. Prediction of pathological upgrading at radical prostatectomy in prostate cancer eligible for active surveillance: a texture features and machine learning-based analysis of apparent diffusion coefficient maps[J]. Front Oncol, 2021, 10: 604266. [19] Zhang Y, Zhu YF, Zhang K, et al. Invasive ductal breast cancer: preoperative predict Ki-67 index based on radiomics of ADC maps[J]. Radiol Med, 2020, 125(2): 109-16. [20] Yu HW, Meng XQ, Chen H, et al. Predicting the level of tumorinfiltrating lymphocytes in patients with breast cancer: usefulness of mammographic radiomics features[J]. Front Oncol, 2021, 11((3)): 628577. [21] 俞飞虹, 王剑翔, 邓晶, 等. 基于多中心超声影像组学特征预测三阴性乳腺癌无病生存期的应用价值[J]. 中华超声影像学杂志, 2021(6): 519-25. [22] 刘瑾瑾, 王鸿凤, 崔浩, 等. 常规超声及剪切波弹性成像在三阴性乳腺癌诊断中的应用价值[J]. 中华超声影像学杂志, 2018, 27(11): 968-72. [23] 李佳伟, 时兆婷, 侯英敏, 等. 浸润性三阴性乳腺癌超声特征的临床、病理学及免疫组织化学基础[J]. 中国癌症杂志, 2019, 29(1): 37-44. https://www.cnki.com.cn/Article/CJFDTOTAL-ZGAZ201901007.htm [24] Tagliafico AS, Piana M, Schenone D, et al. Overview of radiomics in breast cancer diagnosis and prognostication[J]. Breast, 2020, 49: 74-80. [25] Moon WK, Huang YS, Lo CM, et al. Computer-aided diagnosis for distinguishing between triple-negative breast cancer and fibroadenomas based on ultrasound texture features[J]. Med Phys, 2015, 42(6Part1): 3024-35. [26] 王彤, 何萍, 苏畅, 等. 计算机辅助多模态融合超声诊断乳腺良恶性肿瘤[J]. 中国医学影像技术, 2021, 37(8): 1210-3. https://www.cnki.com.cn/Article/CJFDTOTAL-ZYXX202108027.htm -
点击查看大图
计量
- 文章访问数: 1320
- HTML全文浏览量: 389
- PDF下载量: 64
- 被引次数: 0
