分享:
分享到微信朋友圈
X
临床研究
磁共振高分辨率延迟期的可解释性机器学习模型术前预测浸润性乳腺癌组织学分级
匡静 黄松涛 黄小华 胡云涛

Cite this article as: KUANG J, HUANG S T, HUANG X H, et al. Interpretable machine learning model for predicting preoperative histological grade of invasive breast cancer based on high resolution delay period of magnetic resonance imaging[J]. Chin J Magn Reson Imaging, 2025, 16(5): 164-169, 216.本文引用格式:匡静, 黄松涛, 黄小华, 等. 磁共振高分辨率延迟期的可解释性机器学习模型术前预测浸润性乳腺癌组织学分级[J]. 磁共振成像, 2025, 16(5): 164-169, 216. DOI:10.12015/issn.1674-8034.2025.05.025.


[摘要] 目的 探讨基于磁共振高分辨率增强延迟期图像的夏普利加性解释(Shapley additive explanations, SHAP)机器学习模型术前预测非特殊型浸润性乳腺癌组织分级的价值。材料与方法 回顾性收集2019年1月至2023年12月154例非特殊型浸润性乳腺癌患者的临床-病理-影像学资料,根据病理活检结果将Ⅰ级、Ⅱ级纳入低级别组和Ⅲ级纳入高级别组。按7∶3比例随机分为训练组(n=107)和验证组(n=47)。利用3D slicer勾画病灶边缘并提取影像组学特征。通过多因素分析筛选影像组学特征。通过随机森林(random forest, RF)、logistic回归建立影像组学特征模型,采用logistic回归建立临床模型、影像学模型,以及影像学-临床-影像组学特征联合模型。采用受试者工作特征(receiver operating characteristic, ROC)曲线下面积AUC(area under the curve, AUC)、准确率评估模型的效能,模型比较采用DeLong检验。通过SHAP分析可视化特征在模型中的贡献度和重要性。结果 孕激素受体(progesterone receptor, PR)、肿瘤边界、细胞增殖因子(Ki-67)、雌激素受体(estrogen receptor, ER)在低级别组和高级别组之间差异有统计学意义(P<0.05)。基于影像学-临床-影像组学特征的联合模型术前预测浸润性乳腺癌组织分级的AUC值最高,在训练组和验证组的AUC分别为0.807(95% CI:0.723~0.891)、0.890(95% CI:0.795~0.984)。两种独立的影像组学特征模型中,logistic影像组学模型无明显过拟合,在训练组和验证组的AUC分别为0.750(95% CI:0.655~0.846)、0.801(95% CI:0.667~0.936)。临床模型、影像学模型在训练组的AUC分别为0.661(95% CI:0.551~0.771),0.600(95% CI:0.493~0.706),在验证组的AUC分别为0.789(95% CI:0.645~0.933),0.708(95% CI:0.565~0.850)。结论 联合模型术前预测非特殊型浸润性乳腺癌分级的效能较好,可为临床术前治疗乳腺癌提供一定指导。
[Abstract] Objective To explore the value of Shapley additive explanations (SHAP) interpretable machine learning models based on high-resolution enhanced delayed-phase magnetic resonance imaging in preoperatively predicting histologic grade of non-special type invasive breast cancer.Materials and Methods Retrospectively collected the clinical-pathological-imaging data of 154 patients with invasive breast carcinoma of no special type from January 2019 to December 2023. Based on pathological biopsy results, Grade Ⅰ and Ⅱ were classified into the low-grade group, while Grade Ⅲ was classified into the high-grade group. They were randomly divided into a training group of 107 cases and a validation group of 47 cases in a 7∶3 ratio. 3D Slicer was used to delineate the lesion edges and extract radiomics features. Features were screened through multifactorial analysis. Radiomics feature models were established using Random forest (RF) and logistic regression, while clinical models, radiology models, and a combined radiology-clinical-radiomics feature model were developed using logistic regression. The performance of the model was evaluated using the area under the receiver operating characteristic curve (AUC) and accuracy, while model comparison was conducted using DeLong test. SHAP analysis was used to visualize the contribution and importance of features in the model.Results There were significant differences in progesterone receptor (PR), tumor boundary, Ki-67 and estrogen receptor (ER) between low-grade group and high-grade group (P < 0.05). The AUC of the combined model based on radiology-clinical-radiomics features for preoperative prediction of the histological grade of invasive breast cancer was relatively good, with AUC values of 0.807 (95% CI: 0.723 to 0.891) in the training group and 0.890 (95% CI: 0.795 to 0.984) in the validation group. Among the two independent radiomics feature models, the logistic radiomics model showed no obvious overfitting, with AUC values of 0.750 (95% CI: 0.655 to 0.846) in the training group and 0.801 (95% CI: 0.667 to 0.936) in the validation group. The AUC of the clinical model and the radiology model in the training group were 0.661 (95% CI: 0.551 to 0.771) and 0.600 (95% CI: 0.493 to 0.706), respectively, and in the validation group were 0.789 (95% CI: 0.645 to 0.933) and 0.708 (95% CI: 0.565 to 0.850), respectively.Conclusions The joint model showed good efficacy in preoperatively predicting histologic grade of non-special type invasive breast cancer, providing guidance for preoperative treatment of breast cancer patients in clinical practice.
[关键词] 磁共振成像;可解释性机器学习;高分辨率延迟期成像;浸润性乳腺癌;分级
[Keywords] magnetic resonance imaging;explainable machine learning;high-resolution delayed-phase imaging;invasive breast cancer;grading

匡静 1   黄松涛 1   黄小华 2*   胡云涛 3  

1 四川大学华西广安医院广安市人民医院放射科,广安 638500

2 川北医学院附属医院放射科,南充 637000

3 四川省肿瘤医院放射科,成都 610000

通信作者:黄小华,E-mail: 15082797553@163.com

作者贡献声明:黄小华设计本研究的方案,对稿件重要内容进行了修改,获得南充市校科技战略合作项目资助;匡静起草和撰写稿件,获取、分析和解释本研究的数据;黄松涛设计本研究的方案,分析本研究数据,对稿件重要内容进行了修改;胡云涛分析、解释本研究的数据,对稿件重要内容进行了修改。全体作者都同意发表最后的修改稿,同意对本研究的所有方面负责,确保本研究的准确性和诚信。


基金项目: 南充市校科技战略合作项目 19SXHZ0429
收稿日期:2024-11-11
接受日期:2025-05-09
中图分类号:R445.2  R737.9 
文献标识码:A
DOI: 10.12015/issn.1674-8034.2025.05.025
本文引用格式:匡静, 黄松涛, 黄小华, 等. 磁共振高分辨率延迟期的可解释性机器学习模型术前预测浸润性乳腺癌组织学分级[J]. 磁共振成像, 2025, 16(5): 164-169, 216. DOI:10.12015/issn.1674-8034.2025.05.025.

0 引言

       近年来乳腺癌的发病率逐年增长并有年轻化的趋势[1]。低分化的乳腺癌肿瘤细胞不仅具有更高的恶性程度和侵袭性,还有生长快、易转移、对治疗反应差等特点[2],所以此类患者往往面临更严峻的治疗挑战。因此,准确对乳腺癌进行组织分级对确定治疗方案和预后至关重要[3, 4]。目前病理活检是确定乳腺癌组织分级的金标准,但其具有有创性和延迟性等缺点,不能及时为临床治疗提供信息。当前人工智能日益发展的背景下,影像组学在疾病诊断、分级及疗效评估中得到了广泛的应用[5, 6, 7]。将乳腺癌患者的MRI、超声、X线影像数据与机器学习模型相结合,就可以将其应用在乳腺癌的诊断、分级、疗效评估中[8, 9, 10]。杜小萌等[11]从142例乳腺癌患者的动态增强图像(dynamic contrast enhanced, DCE)和弥散加权成像(diffusion weighted imaging, DWI)图像中提取影像组学特征,通过logistic回归建模发现基于DCE-DWI的联合模型效能比独立的DCE影像组学模型和DWI影像组学模型高,在验证组中AUC为0.76。郭亚欣等[12]基于瘤内、瘤周3 mm和瘤周5 mm、瘤周+瘤内的靶区上建立的5种预测浸润性乳腺癌组织分级的影像组学模型预测浸润性乳腺癌组织分级的AUC均高于0.7。夏普利加性解释(Shapley additive explanations, SHAP)是一种解释多种机器学习模型输出的博弈论方法,可以通过SHAP值将特征对模型的贡献可视化。SHAP在乳腺癌的分子分型诊断、新辅助放化疗预后预测等中有广泛应用[13, 14, 15]。然而目前未见有基于磁共振高分辨率图像的影像组学术前预测浸润性乳腺癌组织学分级的SHAP分析研究。因此本研究旨在探讨基于磁共振高分辨率增强延迟期图像的SHAP机器学习模型术前预测非特殊型浸润性乳腺癌组织分级的价值,以期为临床个性化治疗浸润性乳腺癌提供新方法。

1 材料与方法

1.1 研究对象

       本研究遵守《赫尔辛基宣言》,经过广安市人民医院医学伦理委员会批准,免除受试者知情同意(批准文号:2025年审011号)。回顾性收集2019年1月至2023年12月于广安市人民医院行术前乳腺MRI检查的患者,根据Nottingham组织学分级结果分为I级、Ⅱ级、Ⅲ级,本研究将Ⅰ/Ⅱ级归为低级别组,Ⅲ级归为高级别组进行二分类。以7∶3的比例分为训练组和验证组。纳入标准:(1)具有组织学分级结果和病理免疫组化结果[16];(2)术前行乳腺MRI检查,且检查时间与手术间隔不超过2周;(3)均为肿块型非特殊型浸润性乳腺癌;(4)检查前未经任何临床治疗。排除标准:(1)图像受伪影干扰不清晰;(2)临床资料缺失者;(3)检查前行乳腺穿刺;(4)有乳腺手术史或乳腺其他疾病。

1.2 扫描参数

       采用德国西门子3.0 T skyra SIEMENS磁共振扫描仪检查,16通道双侧乳腺专用相控阵列线圈。患者双侧上肢举过头顶,头先进俯卧位,双侧乳腺位于线圈中心并悬空。扫描范围包括两侧完整乳腺。扫描序列与参数:

       轴位T1WI高分辨率增强延迟序列:视野380 mm×380 mm,间距20%,层厚0.8 mm,重复时间8.73 ms,回波时间4.32 ms,层内分辨率336×448,层间分辨率0.96 mm,相位分辨率407×448。增强采用高压注射器(Empower,BRACCO,意大利),流速:2.5 mL/s,注射剂量:0.1 mmol/kg(莫迪司,钆贝葡胺,上海博莱科信谊药业有限公司,中国),并于打药后9 min 30 s后开始扫描。

1.3 病理信息

       根据免疫组化结果将雌激素受体(estrogen receptor, ER)、孕激素受体(progesterone receptor, PR)阳性细胞大于1%定义为阳性。根据中国抗癌协会乳腺专委会推荐,本研究将细胞增殖因子(Ki-67)≥20%定义为高表达组,Ki-67<20%低表达组。人表皮生长因子受体-2(human epidermal growth factor receptor 2, HER-2)“+”认为阴性,“++”加做荧光杂交确定HER-2基因扩增情况,“+++”认为阳性[17, 18]

1.4 影像学资料

       由两名放射科医师(工作5年的主治医师和工作10年的副主任医师)根据《中国抗癌协会乳腺癌诊治指南与规范(2024年版)》[19],在双盲下对患者的肿瘤位置、肿瘤大小、背景实质强化方式(轻度、中度-重度强化)、腺体类型、肿瘤边界(规则、不规则、毛刺)、强化方式(均匀、不均匀、边缘强化)、动态曲线类型(流入型、流出型、平台型)、是否有瘤周水肿进行评估,若两名医师结论不一致通过协商解决。

1.5 特征提取

       由两名具有5年诊断经验的放射科主治医师使用开源的3Dslicer(版本号5.0.3,National Alliance for Medical Imaging Computing,美国)软件,在乳腺癌患者MRI高分辨率增强延迟期图像上逐层对病灶进行勾画(图1),并通过重采样对图像进行预处理,x,y,z参数分别1,1,1。随后从靶区中提取影像组学特征(包括:灰度共生矩阵、灰度运行长度矩阵、一阶特征、灰度依赖矩阵、形状、邻域灰度差矩阵、灰度大小区域矩阵)。

图1  感兴趣区勾画图。女,57岁,非特殊型浸润性乳腺癌。手动逐层勾画病灶三维边缘,勾画时避开邻近血管、正常腺体组织(蓝色箭头标识处为病灶)。
Fig. 1  Outline of the area of interest. Female, 57 years old, non-special type invasive breast cancer. Manually delineate the three-dimensional margins of the lesion layer by layer, avoiding adjacent blood vessels and normal glands during delineation (The blue arrow indicates the location of the lesion).

1.6 特征降维

       通过组内相关系数(intra-class correlation coefficient, ICC)比较两名医师提取特征的一致性,本研究中ICC≥0.75的特征认为一致性较好,并保留ICC≥0.75的特征。

1.7 模型的建立及可视化

       将ICC后的特征进行Z-score标准化。随后通过最小绝对收缩和选择算子进行降维(least absolute shrinkage and selection operator, LASSO)。logistic回归算法建立临床模型、MRI影像学模型、基于临床-影像学-影像组学特征的联合模型。通过随机森林(random forest, RF)和logistic回归建立独立的影像组学模型。采用受试者工作特征曲线下面积(area under the curve, AUC)、准确率来评估各模型的效能,使用校正曲线评价模型拟合度。模型之间的比较采用DeLong检验。采用公式(1)计算出各患者的Radsocre分数。随后采用列线图建立可视化模型,SHAP分析特征在模型中的贡献度及重要性。

1.8 统计学分析

       统计学分析采用SPSS(版本23.0,IBM公司,美国)和R语言(4.4.0 版本,R Development Core Team,新西兰)软件。对于符合正态分布和方差齐性的计量资料采用两个独立样本的t检验,否则使用非参数检验比较组间差异。计数资料采用卡方检验比较组间差异,理论数T<5但T≥1,并且n≥40时用连续性校正的卡方检验,理论数T<1或n<40则用Fisher's检验。P<0.05认为差异有统计学意义。符合正态分布的计量资料采用均值±标准差表示,不符合正态分布的计量资料采用中位数(上下四分位数)表示,计数资料采用频数(百分比)表示。

1.9 样本量估算

       根据机器学习模型中样本量估算的经验法则[20, 21],采用公式(2)估算模型所需样本量:

       其中,n代表降维后所剩特征数,p代表模型预期的分类错误率。由于本研究是二分类问题,随机猜测的错误率为50%。因此本研究模型p取值0.5。

2 结果

2.1 样本量计算结果

       根据本研究通过LASSO降维后,采取最简洁模型原则,只剩余7个影像组学特征(n=7),模型精确度最少50%(p=0.5),估算样本量至少为140例。

2.2 一般资料

       根据纳入排除标准收集154例经病理证实的非特殊型浸润性乳腺癌患者,其中NottinghamⅠ级15例、Ⅱ级93例、Ⅲ级46例。训练组107例,验证组47例。

       年龄、腋窝淋巴结转移、月经状态、HER-2在高级别组与低级别组之间差异无统计学意义(P>0.05)。ER、PR、Ki-67的表达状态在两组之间差异有统计学意义(P<0.05)(表1)。癌胚抗原、糖类抗原19-9、糖类抗原125、糖类抗原153在两组之间差异无统计学意义(P>0.05)(表2)。肿瘤位置、腺体组织类型、背景强化、肿瘤大小、瘤周水肿、强化方式、TIC类型在两组之间差异无统计学意义(P>0.05);肿瘤边界在两组之间差异有统计学意义(P<0.05)(表3)。

表1  一般资料
Tab. 1  General information
表2  低级别组和高级别组的生化资料
Tab. 2  Biochemical data of low and high level groups
表3  低级别组和高级别组的影像学资料
Tab. 3  Radiology data of low grade group and high grade group

2.3 特征降维

       从154例患者高分辨率增强延迟期图像中提取107个影像组学特征,通过ICC后剩余81特征。LASSO降维后剩余7个最佳影像组学特征(图2)。

图2  套索算法特征选择图。2A:图中曲线代表特征的变化轨迹,黑色垂直线定义了λ的最佳值为0.018;2B:左侧和右侧的虚线分别代表二项式偏差最小模型原则和距离二项式偏差最小值的一个标准差范围内的最简洁模型原则,本研究中采用最简洁模型原则。
Fig. 2  Feature selection map of the LASSO algorithm. 2A: The curves in the figure represent the trajectory of feature changes, and the vertical black line defines the optimal value of λ as 0.018. 2B: The dashed lines on the left and right represent the minimum binomial deviation model principle and the simplest model principle within one standard deviation range from the minimum binomial deviation, respectively. In this study, the simplest model pr-inciple is adopted.

2.4 模型的效能

       临床模型、影像学模型、联合模型在训练组的AUC分别为0.661、0.600、0.807,准确率分别为70.1%、70.1%、77.6%。临床模型、影像学模型、联合模型在验证组的AUC分别为0.789、0.708、0.890,准确率分别为70.2%、70.2%、85.1%。RF影像组学特征模型、logistic影像组学特征模型在训练组的AUC分别为0.979、0.750,准确率分别为87.9%、73.8%。RF影像组学特征模型、logistic影像组学特征模型在验证组的AUC分别为0.803、0.801,准确率分别为70.2%、70.2%(表4)。

       联合模型在验证组中的预测效能优于临床模型及影像学模型,差异有统计学意义(联合模型与临床模型相比较,P=0.038;联合模型与影像学模型比较,P=0.018)。校正曲线显示联合模型拟合度较好(图3)。

图3  联合模型校准曲线。蓝色实线代表联合模型校准曲线,黑色虚线代表理想线,蓝色实线越接近黑色虚线代表模型的预测概率与实际发生概率相匹配。
Fig. 3  Joint model calibration curve. The solid blue line represents the joint model calibration curve, the black dashed line represents the ideal line, the closer the blue solid line is to the black dashed line, the more likely the model's predicted probability matches the actual occurrence probability.
表4  五种模型的效能
Tab. 4  Performance of the five models

2.5 模型可视化

       将基于5个特征构建的联合模型可视化为列线图(图4)。SHAP值排序显示,患者的Radscore值对联合模型的预测效能贡献度最大,其次分别是ER表达状态、肿瘤边缘、PR表达状态、Ki-67表达状态(图5)。

图4  联合模型的列线图。影像组学评分和ER、PR、Ki-67、肿瘤边缘的值垂直于Points轴,可获得各项所得评分,将所得评分垂直于Total Points,可得出乳腺癌组织分级的风险概率。ER:雌激素受体;PR:孕激素受体;Ki-67:肿瘤增殖抗原。
Fig. 4  Nomogram of joint model. The values of the Radscore and ER, PR, Ki-67, and tumor boundary are perpendicular to the Points axis, and the resulting scores can be obtained. The risk probability of breast cancer tissue grading can be obtained by perpendicularly aligning the resulting scores with the Total Points axis. ER: estrogen receptor; PR: progesterone receptor; Ki-67: tumor proliferation antigen.
图5  联合模型的夏普利加性解释(SHAP)图。5A:根据SHAP值排序的特征重要性。5B:根据SHAP值特征贡献度的热图。
Fig. 5  Shapley additive explanations (SHAP) of joint model. 5A: Importance of features sorted by SHAP value. 5B: Heat map of characteristic contribution according to SHAP value.

3 讨论

       本研究中建立了临床模型、影像学模型、独立的影像组学特征模型和影像学特征-临床特征-影像组学特征联合模型,并对这几种模型的预测效能进行了比较,发现基于影像学特征-临床特征-影像组学特征的联合模型术前预测浸润性乳腺癌组织学分级的效能较好且稳定,能术前评估乳腺癌患者肿瘤异质性,有助于临床制订更加精准的监测和随访计划。

3.1 影像组学在乳腺癌组织学分级中的现状

       HUANG等[22]研究从常规T2WI压脂序列上建立了术前预测乳腺癌分级的影像组学模型,其AUC为0.812。韩剑剑等[23]通过X线钼靶图像的纹理特征和临床特征建立术前预测乳腺癌组织分级的联合模型,发现联合模型的效果较好,验证组及外部测试组的AUC均>0.8。MAO等[24]从低能图像和减影图像中分别获取了17个和11个影像组学特征,并将两种图像的特征联合建模,联合模型术前预测浸润性乳腺癌的亦均≥0.8。钼靶是乳腺癌筛查的重要手段之一,但钼靶的软组织对比度不及MRI。MRI有多参数和功能成像等特点,但MRI平扫无法评估病灶的血供。虽然动态增强的时间分辨率高扫描速度快,可以显示且评估病灶的血供,但其空间分辨率不足,不能清晰地显示病灶浸润范围及乳腺导管的走行等[25, 26, 27]。由于高分辨率延迟期图像具有空间分辨率高、软组织对比度高、三维显示等优点,可清晰显示肿瘤的边缘、及肿瘤与周围血管的关系。因此本研究在浸润性乳腺癌患者的高分辨率延迟期图像上提取了影像组学特征。

3.2 临床-影像学特征分析

       本研究中通过分析浸润性乳腺癌患者的临床-病理-生化特征发现ER阳性和PR阳性更容易出现在低级别组,Ki-67高表达更容易出现在高级别组,此结果跟WU等[28]研究结果一致,导致此结果的原因可能是不同年龄阶段的雌激素和孕激素水平不同,且高级别组的乳腺癌恶性程度更高,细胞增殖速度更快。韩剑剑等[23]研究发现腋窝淋巴结转移在Ⅰ/Ⅱ类组和Ⅲ类组之间差异有统计学意义,而本研究中腋窝淋巴结转移在低级别组和高级别组中差异无统计学意义,导致此结果的原因可能是本研究的样本量较少。此外,本研究还分析了浸润性乳腺癌患者的乳腺MRI影像学特征,发现肿瘤边缘不规则更容易出现在高级别组,毛刺更容易出现在低级别组,导致此结果的原因可能是低级别的乳腺癌相对于高级别的乳腺癌生长缓慢,侵袭性较弱,周围纤维组织增生较多,此结果与梁园园等[29]研究结果一致。

3.3 主要研究成果分析

       本研究通过特征降维后剩余7个(伸长率、最小轴长、最大2D直径行、相关信息测度、反转方差、低灰度区域强调、区域熵)最佳影像组学特征。其中伸长率、最小轴长、最大2D直径行属于形状特征组,其可以用来描述病变的大小、形状、结构等信息[30]。相关信息测度、反转方差属于灰度共生矩阵特征组,其通过灰度的空间相关性来反映纹理[31]。低灰度区域强调、区域熵属于灰度区域大小矩阵特征组,其反映了二维区域内的像素值连续性[32, 33]

       本研究通过临床特征、MRI影像学特征、影像组学特征建立了五种模型。在独立的影像组学特征模型中,RF模型的预测效能略高于logistic回归模型,但RF模型表现出明显的过拟合,这可能是由于RF模型更适用于大样本研究,而logistic回归模型没有过拟合。因此本研究采用了logistics回归建立了基于影像学特征-临床特征-影像组学特征的联合模型,并将其与临床模型、影像学模型进行了比较,发现联合模型相较于临床模型和影像学模型具有较高的预测效能,其在验证组的AUC和准确率分别高达0.890、85.1%。此结果表明联合模型可有效术前预测非特殊型浸润性乳腺癌组织学分级,且预测效能较好。

       虽然目前机器学习在疾病各方面的研究较广,但受图像采集设备、模型算法,以及影像组学特征无法解释相应的生物学行为等影响[34, 35, 36],导致机器学习尚未在临床大范围地应用。而本研究中通过SHAP全局分析解释了各特征在联合模型中的贡献度和重要性,发现Radscore值越高对模型正向输出影响越大,ER阴性、PR阳性、Ki-67阳性是浸润性高级别乳腺癌的高危因素,肿瘤边界毛刺特征对模型的负向输出影响越大[37, 38, 39]。此结果表明利用SHAP分析可进一步帮助临床工作人员理解模型的分类机制,推动机器学习在临床中的应用。

3.4 本研究的局限性

       (1)未与乳腺MRI常规参数图像的特征进行比较,成像参数单一;(2)未采用多算法建模并进行比较;(3)样本量相对较少,本研究将Ⅰ类级别和Ⅱ类级别归为了一组,未进行三分类。以上局限性可能会限制模型的泛化能力,且无法对Ⅰ、Ⅱ、Ⅲ类级别精确分类。因此,在未来研究中将进一步扩大样本量,并使用多参数、多算法比较不同模型之间的效能。

4 结论

       综上所述,基于磁共振高分辨率延迟期图像的SHAP释联合模型可术前预测非特殊型浸润性乳腺癌组织分级,并可为临床个性化治疗乳腺癌提供一定指导。

[1]
WILKINSON L, GATHANI T. Understanding breast cancer as a global health concern[J/OL]. Br J Radiol, 2022, 95(1130): 20211033 [2024-11-10]. https://pubmed.ncbi.nlm.nih.gov/34905391/. DOI: 10.1259/bjr.20211033.
[2]
PEART O. Metastatic breast cancer[J/OL]. Radiol Technol, 2017, 88(5): 519M-539M. [2024-11-10]. https://pubmed.ncbi.nlm.nih.gov/28500107/.
[3]
RAKHA E, TOSS M, QUINN C. Specific cell differentiation in breast cancer: a basis for histological classification[J]. J Clin Pathol, 2022, 75(2): 76-84. DOI: 10.1136/jclinpath-2021-207487.
[4]
TESCH M E, PARTRIDGE A H. Treatment of breast cancer in young adults[J/OL]. Am Soc Clin Oncol Educ Book, 2022, 42: 1-12 [2024-11-10]. https://pubmed.ncbi.nlm.nih.gov/35580291/. DOI: 10.1200/EDBK_360970.
[5]
KOVAČEVIĆ L, ŠTAJDUHAR A, STEMBERGER K, et al. Breast cancer surrogate subtype classification using pretreatment multi-phase dynamic contrast-enhanced magnetic resonance imaging radiomics: a retrospective single-center study[J/OL]. J Pers Med, 2023, 13(7): 1150 [2024-11-10]. https://pubmed.ncbi.nlm.nih.gov/37511763/. DOI: 10.3390/jpm13071150.
[6]
SHENG W Y, XIA S L, WANG Y R, et al. Invasive ductal breast cancer molecular subtype prediction by MRI radiomic and clinical features based on machine learning[J/OL]. Front Oncol, 2022, 12: 964605 [2024-11-10]. https://pubmed.ncbi.nlm.nih.gov/36172153/. DOI: 10.3389/fonc.2022.964605.
[7]
TABNAK P, HAJIESMAILPOOR Z, BARADARAN B, et al. MRI-based radiomics methods for predicting ki-67 expression in breast cancer: a systematic review and meta-analysis[J]. Acad Radiol, 2024, 31(3): 763-787. DOI: 10.1016/j.acra.2023.10.010.
[8]
LI Y F, LONG W X, ZHOU H D, et al. Revolutionizing breast cancer ki-67 diagnosis: ultrasound radiomics and fully connected neural networks (FCNN) combination method[J]. Breast Cancer Res Treat, 2024, 207(2): 453-468. DOI: 10.1007/s10549-024-07375-x.
[9]
LIU S S, DU S Y, GAO S, et al. A delta-radiomic lymph node model using dynamic contrast enhanced MRI for the early prediction of axillary response after neoadjuvant chemotherapy in breast cancer patients[J/OL]. BMC Cancer, 2023, 23(1): 15 [2024-11-10]. https://pubmed.ncbi.nlm.nih.gov/36604679/. DOI: 10.1186/s12885-022-10496-5.
[10]
张丁懿, 黄小华, 沈梦伊, 等. MRI影像组学预测非特殊型浸润性乳腺癌分子分型的价值[J]. 磁共振成像, 2024, 15(3): 100-106. DOI: 10.12015/issn.1674-8034.2024.03.017.
ZHANG D Y, HUANG X H, SHEN M Y, et al. Value of MRI radiomics in predicting molecular subtypes of invasive breast carcinoma of no special type[J]. Chin J Magn Reson Imag, 2024, 15(3): 100-106. DOI: 10.12015/issn.1674-8034.2024.03.017.
[11]
杜小萌, 李艳玲, 曹胜男, 等. DCE-MRI和DWI影像组学特征预测浸润性乳腺癌病理分级的价值[J]. 临床放射学杂志, 2022, 41(9): 1641-1644. DOI: 10.13437/j.cnki.jcr.2022.09.002.
DU X M, LI Y L, CAO S N, et al. The value of DCE-MRI and DWI radiomics features in predicting the pathological grading of invasive breast cancer[J]. J Clin Radiol, 2022, 41(9): 1641-1644. DOI: 10.13437/j.cnki.jcr.2022.09.002.
[12]
郭亚欣, 王贇霞, 尚怡研, 等. 弥散加权成像瘤内及瘤周影像组学预测乳腺癌组织学分级[J]. 中国介入影像与治疗学, 2024, 21(3): 160-165. DOI: 10.13929/j.issn.1672-8475.2024.03.007.
GUO Y X, WANG Y X, SHANG Y Y, et al. Intratumoral and peritumoral radiomics based on diffusion weighted imaging for predicting histological grade of breast cancer[J]. Chin J Interv Imag Ther, 2024, 21(3): 160-165. DOI: 10.13929/j.issn.1672-8475.2024.03.007.
[13]
陈鑫, 陈小波, 梁演婷, 等. 基于MRI影像组学的可解释性机器学习模型预测乳腺癌新辅助治疗病理完全缓解[J]. 临床放射学杂志, 2024, 43(10): 1666-1672. DOI: 10.13437/j.cnki.jcr.2024.10.027.
CHEN X, CHEN X B, LIANG Y T, et al. Interpretable machine learning model basedon MRI radiomics for predicting pathological complete response to neoadjuvant therapy in breast cancer[J]. J Clin Radiol, 2024, 43(10): 1666-1672. DOI: 10.13437/j.cnki.jcr.2024.10.027.
[14]
舒予静, 何子龙, 曾伟雄, 等. 基于双模态乳腺影像特征的HER2表达状态多分类机器学习预测模型构建及可解释性分析研究[J]. 临床放射学杂志, 2024, 43(8): 1310-1316. DOI: 10.13437/j.cnki.jcr.2024.08.016.
SHU Y J, HE Z L, ZENG W X, et al. Multi-class machine learning predictive model for HER2 expression status based on dual-modal breast imaging features[J]. J Clin Radiol, 2024, 43(8): 1310-1316. DOI: 10.13437/j.cnki.jcr.2024.08.016.
[15]
MA M W, LIU R Y, WEN C J, et al. Predicting the molecular subtype of breast cancer and identifying interpretable imaging features using machine learning algorithms[J]. Eur Radiol, 2022, 32(3): 1652-1662. DOI: 10.1007/s00330-021-08271-4.
[16]
MEI F, LIU J Y, XUE W C. Histological grading of invasive breast carcinoma: Nottingham histological grading system[J]. Zhonghua Bing Li Xue Za Zhi, 2019, 48(8): 659-664. DOI: 10.3760/cma.j.issn.0529-5807.2019.08.019.
[17]
张丽, 黄小华, 沈梦伊, 等. 基于不同机器学习算法的影像组学模型预测浸润性乳腺癌Ki-67表达的价值[J]. 中国医学计算机成像杂志, 2024, 30(1): 39-44. DOI: 10.19627/j.cnki.cn31-1700/th.2024.01.016.
ZHANG L, HUANG X H, SHEN M Y, et al. The value of radiomics models based on different machine learning in predicting ki-67 expression in invasive breast cancer[J]. Chin Comput Med Imag, 2024, 30(1): 39-44. DOI: 10.19627/j.cnki.cn31-1700/th.2024.01.016.
[18]
朱浩雨, 陈基明, 葛亚琼, 等. MRI影像组学术前预测浸润性乳腺癌脉管侵犯的价值[J]. 中国医学影像学杂志, 2020, 28(11): 825-830. DOI: 10.3969/j.issn.1005-5185.2020.11.007.
ZHU H Y, CHEN J M, GE Y Q, et al. Preoperative prediction of lymphovascular invasion in invasive breast cancer with MRI radiomics[J]. Chin J Med Imag, 2020, 28(11): 825-830. DOI: 10.3969/j.issn.1005-5185.2020.11.007.
[19]
中国抗癌协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诊治指南与规范(2024年版)[J]. 中国癌症杂志, 2023, 33(12): 1092-1187. DOI: 10.19401/j.cnki.1007-3639.2023.12.004.
Society of Breast Cancer China Anti-Cancer Association, Breast Oncology Group of the Oncology Branch of the Chinese Medical Association. Guidelines for breast cancer diagnosis and treatment by China Anti-cancer Association(2024 edition)[J]. China Oncol, 2023, 33(12): 1092-1187. DOI: 10.19401/j.cnki.1007-3639.2023.12.004.
[20]
PEDUZZI P, CONCATO J, KEMPER E, et al. A simulation study of the number of events per variable in logistic regression analysis[J]. J Clin Epidemiol, 1996, 49(12): 1373-1379. DOI: 10.1016/s0895-4356(96)00236-3.
[21]
RILEY R D, SNELL K I, ENSOR J, et al. Minimum sample size for developing a multivariable prediction model: PART II-binary and time-to-event outcomes[J]. Stat Med, 2019, 38(7): 1276-1296. DOI: 10.1002/sim.7992.
[22]
HUANG Y, CHENG Z, HUANG X, et al. Preoperative evaluation of histologic grade in invasive breast cancer with T2W-MRI based radiomics signature[J]. Zhong Nan da Xue Xue Bao Yi Xue Ban, 2019, 44(3): 285-289. DOI: 10.11817/j.issn.1672-7347.2019.03.009.
[23]
韩剑剑, 马培旗, 王小雷, 等. 基于数字化乳腺X线影像组学列线图预测浸润性乳腺癌组织学分级的多中心研究[J]. 放射学实践, 2024, 39(5): 634-640. DOI: 10.13609/j.cnki.1000-0313.2024.05.012.
HAN J J, MA P Q, WANG X L, et al. Predictive value of histological grading of invasive breast cancer based on digital mammography radiomics nomogram: a multicenter study[J]. Radiol Pract, 2024, 39(5): 634-640. DOI: 10.13609/j.cnki.1000-0313.2024.05.012.
[24]
MAO N, JIAO Z M, DUAN S F, et al. Preoperative prediction of histologic grade in invasive breast cancer by using contrast-enhanced spectral mammography-based radiomics[J]. J Xray Sci Technol, 2021, 29(5): 763-772. DOI: 10.3233/XST-210886.
[25]
BRAMAN N M, ETESAMI M, PRASANNA P, et al. Intratumoral and peritumoral radiomics for the pretreatment prediction of pathological complete response to neoadjuvant chemotherapy based on breast DCE-MRI[J/OL]. Breast Cancer Res, 2017, 19(1): 57 [2024-11-10]. https://pubmed.ncbi.nlm.nih.gov/28521821/. DOI: 10.1186/s13058-017-0846-1.
[26]
CHOI B B. Dynamic contrast enhanced-MRI and diffusion-weighted image as predictors of lymphovascular invasion in node-negative invasive breast cancer[J/OL]. World J Surg Oncol, 2021, 19(1): 76 [2024-11-10]. https://pubmed.ncbi.nlm.nih.gov/33722246/. DOI: 10.1186/s12957-021-02189-3.
[27]
PENG S Y, CHEN L Q, TAO J, et al. Radiomics analysis of multi-phase DCE-MRI in predicting tumor response to neoadjuvant therapy in breast cancer[J/OL]. Diagnostics (Basel), 2021, 11(11): 2086 [2024-11-10]. https://pubmed.ncbi.nlm.nih.gov/34829433/. DOI: 10.3390/diagnostics11112086.
[28]
WU Y T, PAN W X, WANG L X, et al. A comprehensive model based on dynamic contrast-enhanced magnetic resonance imaging can better predict the preoperative histological grade of breast cancer than a radiomics model[J/OL]. Breast Cancer (Dove Med Press), 2023, 15: 709-720 [2024-11-10]. https://pubmed.ncbi.nlm.nih.gov/37873521/. DOI: 10.2147/BCTT.S425996.
[29]
梁园园, 李昌帅, 赵文静, 等. 多模态影像学术前评估浸润性乳腺癌组织学分级[J]. 中国介入影像与治疗学, 2021, 18(10): 617-621. DOI: 10.13929/j.issn.1672-8475.2021.10.010.
LIANG Y Y, LI C S, ZHAO W J, et al. Multimodal imaging for preoperational prediction of histologic classification of invasive breast cancer[J]. Chin J Interv Imag Ther, 2021, 18(10): 617-621. DOI: 10.13929/j.issn.1672-8475.2021.10.010.
[30]
ÖZDEMIR H, AZAMAT S, ÖZDEMIR M S. Can only the shape feature in radiomics help machine learning show that bladder cancer has invaded muscles?[J/OL]. Cureus, 2023, 15(9): e45488 [2024-11-10]. https://pubmed.ncbi.nlm.nih.gov/37859896/. DOI: 10.7759/cureus.45488.
[31]
ELAZAB N, GAB ALLAH W, ELMOGY M. Computer-aided diagnosis system for grading brain tumor using histopathology images based on color and texture features[J/OL]. BMC Med Imaging, 2024, 24(1): 177 [2024-11-10]. https://pubmed.ncbi.nlm.nih.gov/39030508/. DOI: 10.1186/s12880-024-01355-9.
[32]
MAHMOOD U, APTE A P, DEASY J O, et al. Investigating the robustness neighborhood gray tone difference matrix and gray level co-occurrence matrix radiomic features on clinical computed tomography systems using anthropomorphic phantoms: evidence from a multivendor study[J]. J Comput Assist Tomogr, 2017, 41(6): 995-1001. DOI: 10.1097/RCT.0000000000000632.
[33]
CASTELLOTE-HUGUET P, RUIZ-ESPANA S, GALAN-AUGE C, et al. Breast cancer diagnosis using texture and shape features in MRI[J/OL]. Annu Int Conf IEEE Eng Med Biol Soc, 2023, 2023: 1-4 [2024-11-10]. https://pubmed.ncbi.nlm.nih.gov/38083249/. DOI: 10.1109/EMBC40787.2023.10340385.
[34]
冯晓丹, 刘菲菲, 杜瑶, 等. 基于超声影像组学及临床特征的可解释性机器学习模型预测乳腺癌新辅助化疗疗效[J]. 中国超声医学杂志, 2024, 40(3): 278-282. DOI: 10.3969/j.issn.1002-0101.2024.03.010s.
FENG X D, LIU F F, DU Y, et al. Predicting the effect of interpretable machine learning model based on ultrasound radiomics and clinical features on neoadjuvant chemotherapy for breast cancer[J]. Chin J Ultrasound Med, 2024, 40(3): 278-282. DOI: 10.3969/j.issn.1002-0101.2024.03.010s.
[35]
吴宇强, 秦涛, 马晓臣, 等. 实性孤立性肺结节CT影像组学参数测量的可重复性研究[J]. 放射学实践, 2020, 35(9): 1106-1111. DOI: 10.13609/j.cnki.1000-0313.2020.09.006.
WU Y Q, QIN T, MA X C, et al. Study on the reproducibility of CT-based radiomics features in solid solitary pulmonary nodule[J]. Radiol Pract, 2020, 35(9): 1106-1111. DOI: 10.13609/j.cnki.1000-0313.2020.09.006.
[36]
KUMAR V, GU Y H, BASU S, et al. Radiomics: the process and the challenges[J]. Magn Reson Imaging, 2012, 30(9): 1234-1248. DOI: 10.1016/j.mri.2012.06.010.
[37]
ZHANG G S, SHI Y, YIN P P, et al. A machine learning model based on ultrasound image features to assess the risk of sentinel lymph node metastasis in breast cancer patients: Applications of scikit-learn and SHAP[J/OL]. Front Oncol, 2022, 12: 944569 [2024-11-10]. https://pubmed.ncbi.nlm.nih.gov/35957890/. DOI: 10.3389/fonc.2022.944569.
[38]
HU Z, TIAN Z K, WEI X, et al. Machine learning model for predicting axillary lymph node metastasis in clinically node positive breast cancer based on peritumoral ultrasound radiomics and SHAP feature analysis[J]. J Ultrasound Med, 2024, 43(10): 2007-2008. DOI: 10.1002/jum.16520.
[39]
LIU Y, FU Y T, PENG Y D, et al. Clinical decision support tool for breast cancer recurrence prediction using SHAP value in cooperative game theory[J/OL]. Heliyon, 2024, 10(2): e24876 [2024-11-10]. https://pubmed.ncbi.nlm.nih.gov/38312672/. DOI: 10.1016/j.heliyon.2024.e24876.

上一篇 动态对比增强磁共振结合瘤内瘤周影像组学在预测乳腺非肿块强化病灶良恶性中的价值研究
下一篇 基于多参数MR生境成像鉴别Luminal型和非Luminal型乳腺癌
  
诚聘英才 | 广告合作 | 免责声明 | 版权声明
联系电话:010-67113815
京ICP备19028836号-2