分享:
分享到微信朋友圈
X
临床研究
基于Kaiser评分的乳腺MRI肿块诊断预测模型的构建与验证
易熙 王月爱 刘芳 杨宇 陈晓琼 曾禹莉

Cite this article as: YI X, WANG Y A, LIU F, et al. Development and validation of a predictive model for the diagnosis of breast MRI masses based on the Kaiser score[J]. Chin J Magn Reson Imaging, 2023, 14(5): 96-103.本文引用格式:易熙, 王月爱, 刘芳, 等. 基于Kaiser评分的乳腺MRI肿块诊断预测模型的构建与验证[J]. 磁共振成像, 2023, 14(5): 96-103. DOI:10.12015/issn.1674-8034.2023.05.018.


[摘要] 目的 旨在构建一个基于乳腺动态对比增强MRI(dynamic contrast-enhanced MRI, DCE-MRI)Kaiser评分的乳腺肿块诊断预测模型并进行外部验证,用于诊断预测乳腺MRI中肿块的恶性风险。材料与方法 收集2020年5月至2021年3月于湖南省人民医院天心阁院区及2019年9月至2020年12月于湖南省人民医院马王堆院区行术前乳腺DCE-MRI检查并经手术或穿刺病理证实的病灶分别为199个(来自于199名患者)、86个(来自于81名患者,其中5名患者有2个病灶),以天心阁院区数据为训练集,马王堆院区数据为验证集。收集的影像参数包括:乳腺纤维腺体类型、背景实质强化(面积、对称性)、病灶大小、肿块特征(形状、边缘、内部强化特征)、DCE-MRI时间-信号曲线(time-signal intensity curve, TIC)、乳腺水肿情况、最大信号强度投影(maximum intensity projection, MIP)征、附属影像特征(包括:乳头回缩、乳头侵犯、皮肤回缩、皮肤增厚、皮肤侵犯、腋窝淋巴结增大、胸肌侵犯、胸壁侵犯、结构扭曲),并基于Kaiser评分流程图给出Kaiser评分;临床参数包括:年龄、性别、是否伴疼痛、是否可触及肿块、皮肤红肿情况、乳头溢液情况、是否伴橘皮样外观及酒窝征。采用最小绝对收缩和选择算子(least absolute shrinkage and selection operator, LASSO)进行预测变量筛选,多因素logistic回归进行预测模型构建,并以列线图的形式呈现。受试者工作特征(receiver operating characteristic, ROC)曲线、DeLong检验、净重新分类指数(net reclassification index, NRI)及综合判别改善指数(integrated discrimination improvement, IDI)用于比较基于Kaiser评分的乳腺肿块诊断预测模型(以下简称“乳腺肿块诊断模型”)和Kaiser评分的诊断性能;绘制校准曲线以评估乳腺诊断模型的校准度;决策曲线分析(decision curve analysis, DCA)用于评价二者的临床有效性。结果 LASSO回归显示“年龄”“MIP征”及“附属影像特征”是Kaiser评分所用指标之外的有效预测因素;在训练集中,乳腺肿块诊断模型及Kaiser评分的AUC分别为0.944、0.890,差异有统计学意义(P<0.05),在验证集中,乳腺肿块诊断模型及Kaiser评分的AUC分别为0.941、0.874,差异有统计学意义(P<0.05)。DeLong检验及NRI、IDI显示乳腺肿块诊断模型较Kaiser评分对乳腺肿块的诊断性能更好,差异有统计学意义(P<0.05),校准曲线显示乳腺肿块诊断模型的校准度良好;DCA表明乳腺肿块诊断模型具有较高的临床应用价值;结论 基于Kaiser评分的乳腺肿块诊断模型可被用于乳腺肿块恶性概率的术前预测,并且其对乳腺肿块的诊断性能优于经典Kaiser评分。
[Abstract] Objective To construct and externally validate a diagnostic prediction model for breast masses based on the Kaiser score of dynamic contrast-enhanced MRI (DCE-MRI) for diagnostic prediction of the risk of malignancy of masses on breast MRI.Materials and Methods We collected 199 lesions (from 199 patients) and 86 lesions (from 81 patients, including 5 patients with 2 lesions) from the Tianxinge Branch of Hunan Provincial People's Hospital from May 2020 to March 2021 and from the Mawangdui Branch of Hunan Provincial People's Hospital from September 2019 to December 2020, who underwent preoperative breast DCE-MRI and were confirmed by surgical or puncture pathology. Using the data from Tianxinge Branch as the training set and the data from Mawangdui Branch as the validation set. Imaging parameters collected included: the amount of fibroglandular tissue (FGT), background parenchymal enhancement (BPE), lesion size, mass characteristics (shape, margins, internal enhancement features), time-signal intensity curve (TIC), breast edema status, maximum intensity projection (MIP) sign, associated features (including nipple retraction, nipple invasion, skin retraction, skin thickening, skin invasion, axillary lymph node enlargement, pectoral muscle invasion, chest wall invasion, structural distortion), and Kaiser score based on the Kaiser score flow chart. Clinical parameters included age, gender, presence of pain, palpable mass, skin erythema, nipple discharge, orange peel appearance, and dimple sign. The least absolute shrinkage and selection operator (LASSO) was used to select predictor variables. Multivariate logistic regression was used to construct the prediction model, which was presented as a nomogram. The receiver operating characteristic (ROC) curve, DeLong test, net reclassification index (NRI), and integrated discrimination improvement (IDI) were used to compare the diagnostic performance of the Kaiser score-based breast mass diagnostic prediction model (hereinafter referred to as "breast mass diagnostic model") and Kaiser score; calibration curves were plotted to assess the calibration of the breast mass diagnostic model; decision curve analysis (DCA) was used to evaluate the clinical validity of them.Results LASSO regression showed that "age" "MIP sign" and " associated features" were effective predictors in addition to those used in the Kaiser score; In the training set, the AUCs of the breast mass diagnostic model and Kaiser score were 0.944 and 0.890, with statistically significant differences (P<0.05), and in the validation set, the AUCs of the breast mass diagnostic model and Kaiser score were 0.941 and 0.874, with statistically significant differences (P<0.05). Furthermore, NRI and IDI showed that the breast mass diagnostic model had a better diagnostic performance for breast masses than the Kaiser score, and the difference was statistically significant (P<0.05); the calibration curve showed that the breast mass diagnostic model was well calibrated; DCA indicated that the breast mass diagnostic model had high clinical application value.Conclusions The Kaiser score-based diagnostic model for breast masses can be used for preoperative prediction of the probability of malignancy of breast masses. Its diagnostic performance for breast masses is better than the classic Kaiser score.
[关键词] 乳腺癌;乳腺良性病变;乳腺;Kaiser评分;预测模型;列线图;磁共振成像;动态对比增强
[Keywords] breast cancer;benign breast lesions;breast;Kaiser score;prediction model;nomogram;magnetic resonance imaging;dynamic contrast-enhanced

易熙 1, 2   王月爱 1*   刘芳 1   杨宇 3   陈晓琼 1   曾禹莉 2  

1 湖南中医药大学第一附属医院超声影像科,长沙 410007

2 湖南省人民医院(湖南师范大学附属第一医院)放射科,长沙 410016

3 湖南中医药大学第一附属医院放射科,长沙 410007

通信作者:王月爱,E-mail:497375291@qq.com

作者贡献声明:王月爱设计了本研究的方案,对稿件重要内容进行了修改;易熙起草和撰写了稿件,获取、分析或解释本研究的数据;刘芳、杨宇、陈晓琼、曾禹莉获取、分析或解释本研究的数据,对稿件重要内容进行了修改;刘芳获得了湖南省教育厅科学研究项目的资助,杨宇获得了湖南省自然科学基金委员会科卫联合基金资助。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: 湖南省教育厅科学研究项目 21C0236 湖南省自然科学基金委员会科卫联合基金资助项目 2022JJ70114
收稿日期:2022-12-08
接受日期:2023-04-23
中图分类号:R445.2  R737.9 
文献标识码:A
DOI: 10.12015/issn.1674-8034.2023.05.018
本文引用格式:易熙, 王月爱, 刘芳, 等. 基于Kaiser评分的乳腺MRI肿块诊断预测模型的构建与验证[J]. 磁共振成像, 2023, 14(5): 96-103. DOI:10.12015/issn.1674-8034.2023.05.018.

0 前言

       乳腺癌的全球负担正在迅速增加,2020年全球乳腺癌新增患者超过肺癌,达230万例,位居第一位[1, 2]。在中国,乳腺癌仍然是一个重大的公共卫生问题[3],其标化发病率呈快速上升趋势,增幅远高于全球平均水平[4]。随着人口增长和人口老龄化,乳腺癌每年的新增病例和死亡人数将持续增加,需要全球性的努力来改善乳腺癌的诊断、治疗和患者管理,特别是在乳腺癌负担快速上升的发展中国家[5, 6]。在乳腺癌的诊断中,动态对比增强MRI(dynamic contrast-enhanced MRI, DCE-MRI)是主要的影像检查手段之一,对所有类型的乳腺癌都具有较高的敏感性[7, 8, 9],在乳腺癌的早期诊断和疾病范围的准确评估方面,DCE-MRI具有较高的临床应用价值。乳腺影像报告和数据系统(Breast Imaging Reporting and Data System, BI-RADS)为乳腺影像报告提供了标准化的术语词典,但是由于BI-RADS未提供具体的诊断指导策略,在实际运用中,BI-RADS分类诊断较为主观。另外,乳腺良恶性病变之间的影像特征存在一定的重叠,同样给准确诊断带来了挑战。BALTZER等[10, 11]于2013年提出了一种结构化的乳腺MRI诊断工具并将其命名为Kaiser评分,其良好的诊断效能和稳定的可操作性已为国内外多项研究所证实[12, 13, 14]。但是,除Kaiser评分所应用的标准影像诊断参数外,还有一些常见且易获取的诊断参数如年龄、最大信号强度投影(maximum intensity projection, MIP)征等对乳腺病变的MRI诊断同样非常有价值[15, 16, 17]。基于此,对Kaiser评分进行优化以提升其诊断性能是自Kaiser评分被提出以来学界一直比较关心的问题。但相关研究中,新引入的诊断参数存在优化效果不明显或者引入参数在国内应用不普遍的情况,且在引入新参数的方法方面,大部分研究在引入新参数时,未针对各诊断参数在诊断策略中的权重进行统计学分析,对所引入的新参数在诊断策略中的权重分配较为主观。本研究将探讨对乳腺肿块具有诊断价值的其他常见临床、影像诊断参数,并采用多因素logistic回归的方法引入新诊断参数,以此构建基于Kaiser评分的乳腺肿块诊断模型并进行外部验证,并以列线图的形式呈现,以期为乳腺肿块的MRI诊断提供更优的诊断策略。

1 材料与方法

1.1 研究对象

       回顾性分析2020年5月至2021年3月于湖南省人民医院天心阁院区及2019年9月至2020年12月于湖南省人民医院马王堆院区行术前乳腺DCE-MRI检查的患者的临床及影像资料,以天心阁院区数据为训练集,马王堆院区数据为验证集。纳入标准:(1)乳腺DCE-MRI上病灶表现为肿块;(2)病变性质经手术或穿刺病理证实,并且病理结果完整;(3)患者行乳腺DCE-MRI检查前6个月内未进行任何形式的干预诊断措施及治疗(包括穿刺活检、手术、放化疗及内分泌治疗等);(4)截至本次检查时患者未被诊断合并其他恶性肿瘤。排除标准:(1)乳腺DCE-MRI上病灶过小或多发致DCE-MRI所见病灶与穿刺或手术所取病理标本不能确切对应者;(2)临床资料不完整者;(3)MRI图像扫描序列不完整,或因伪影干扰致图像质量不能达到诊断要求者。本研究遵守《赫尔辛基宣言》,经湖南省人民医院(湖南师范大学附属第一医院)医学伦理委员会批准,免除受试者知情同意(批准文号:2022科研伦审第146号)。

1.2 设备和参数

1.2.1 训练集

       MRI扫描使用德国西门子公司Trio Tim 3.0 T MRI仪,乳腺专用相控阵线圈。患者采取俯卧位进行扫描。MRI扫描序列及参数:轴位FSE T1WI平扫(TR 600 ms,TE 13 ms,矩阵340 mm×340 mm),轴位FSE T2WI-STIR平扫(TR 3500 ms,TE 61 ms,FOV 350 mm×192 mm),矢状位T2WI-fs平扫(TR 3500 ms,TE 61 ms,FOV 180 mm×180 mm),轴位DCE-MRI扫描:VIBE压脂序列,TE 4.2 ms,TR 1.6 ms,FOV 350 mm×350 mm,层厚0.9 mm。动态增强序列包括增强前蒙片+7期动态增强,于对比剂注射后15 s开始扫描,57 s/期。对比剂使用钆喷酸葡胺(德国Bayer公司),静脉注射,剂量0.1 mmol/kg,注射速率2.5 mL/s,其后以同样速率注入25 mL生理盐水。

1.2.2 验证集

       MRI扫描使用美国GE公司Signa HDxt 1.5 T MRI仪,乳腺专用相控阵线圈。患者采取俯卧位进行扫描。MRI扫描序列及参数:轴位FSE T1WI平扫(TR 560.0 ms,TE 9.7 ms,FOV 320 mm×224 mm),轴位FSE T2WI-STIR平扫(TR 6140.0 ms,TE 46.5 ms,FOV 320 mm×192 mm),矢状位T2WI-fs平扫(TR 2940.0 ms,TE 87.3 ms,FOV 320 mm×224 mm),轴位DCE-MRI扫描:VIBRANT序列,TR 5.1 ms,TE 2.5 ms,FOV 448 mm×350 mm,层厚0.9 mm。动态增强序列包括增强前蒙片+7期动态增强,于对比剂注射后15 s开始扫描,57 s/期。对比剂使用钆特酸葡胺(中国江苏恒瑞医药公司),静脉注射,剂量0.2 mL/kg,注射速率2.0 mL/s,其后以同样速率注入20 mL生理盐水。

1.3 数据的采集与处理

       通过MRI设备配套的后处理工作站西门子Syngo/GE ADW4.5对图像进行处理,由两名具备乳腺影像诊断专业经验的医师(乳腺DCE-MRI阅片例数≥500例,其中一位为具有23年影像诊断经验的副主任医师,一位为具有9年影像诊断经验的主治医师)在不知道最终病理结果的前提下按照BI-RADS(第5版)标准对相关影像诊断参数进行提取,当结论不一致时,经协商达成一致。在完成影像参数的提取工作后,由上述两位医师在不知道对应的影像参数的前提下分别在医院病历系统和医院病理报告查询系统中提取相关临床参数和病理结果。提取的MRI诊断参数及临床参数见表1表2,训练集及验证集中诊断参数分布情况对比见表3。其中MIP征阳性定义为在减影图像上病灶存在周围血管征或病变侧乳腺血供不对称性增加;附属影像特征包括:乳头回缩、乳头侵犯、皮肤回缩、皮肤增厚、皮肤侵犯、腋窝淋巴结增大、胸肌侵犯、胸壁侵犯、结构扭曲,Kaiser评分根据BALTZER等的Kaiser评分流程图(图1[18]进行。

图1  Kaiser评分流程图。
Fig. 1  Kaiser score flow chart.
表1  训练集基线指标分布情况
Tab. 1  Distribution of baseline metrics in the training set
表2  验证集基线指标分布情况
Tab. 2  Distribution of baseline indicators in the validation set
表3  训练集及验证集中诊断参数分布情况对比
Tab. 3  Comparison of the distribution of diagnostic parameters in the training and validation sets

1.4 统计学分析与预测模型的构建、评估

       使用Empower Stats 4.0版(http://www.empowerstats.com, X&Y Solutions, Inc., Boston, MA, USA)和R统计软件3.5.3版(http://www.R-project.org,The R Foundation)进行统计分析。计量资料以均值±标准差表示,定性资料以频数(百分比)表示。卡方检验或Fisher精确概率法用于定性资料组间差异的比较,t检验或ANOVA检验用于计量资料组间差异的比较。采用LASSO回归进行预测变量筛选;多因素logistic回归构建预测模型,并以列线图的形式呈现。使用受试者工作特征(receiver operating characteristic, ROC)曲线并通过Boostrap法进行500次内部重抽样来评价乳腺肿块诊断模型和Kaiser评分在训练集和验证集中的区分度,通过计算净重新分类指数(net reclassification index, NRI)、综合判别改善指数(integrated discrimination improvement, IDI)及DeLong检验来比较乳腺肿块诊断模型和Kaiser评分的诊断性能,校准曲线用来评估模型的校准度。决策曲线分析(decision curve analysis, DCA)用于评价二者的临床有效性;P<0.05为差异有统计学意义。

2 结果

2.1 患者一般资料

       训练集共计入组病灶199个(来自199名患者),年龄13~89(48.30±14.24)岁,其中女194名(97.49%),男5名(2.51%),良性病灶77个(38.69%),恶性病灶122个(61.31%);验证集共计入组病灶86个(来自81名患者,其中5名患者有2个病灶),年龄17~80(45.56±13.77)岁(同一患者多个病灶按多名患者进行统计),皆为女性,良性病灶56个(65.12%),恶性病灶30个(34.88%);具体病理类型见表4,其中训练集5位男性患者病理类型包括浸润性导管癌2例、乳腺炎1例、良性间叶源性肿瘤1例、男性乳腺发育1例。

表4  训练集与验证集病灶具体病理类型分布
Tab. 4  Distribution of specific pathological types of lesions in the training and validation sets

2.2 预测变量筛选结果

       在训练集基线指标中[除去Kaiser评分流程图中已最终纳入的标准参数“毛刺征”“病灶边缘”“时间-信号曲线(time-signal intensity curve, TIC)”“内部强化模式”和“水肿”]用LASSO回归(通过10重交叉验证筛选lambda,基于lambda.1se即错误均值在最小值的1个标准差范围之内对应的最大lambda)筛选得到三个非零系数的预测指标:年龄、MIP征、附属影像特征(图2)。

图2  LASSO回归系数路径图(2A)与交叉验证曲线(2B)。LASSO:最小绝对收缩和选择算子。
图3  乳腺肿块诊断预测模型列线图。MIP:最大信号强度投影。
Fig. 2  LASSO regression coefficient path diagram (2A) and LASSO regression cross-validation curve (2B). LASSO: least absolute shrinkage and selection operator.
Fig. 3  Nomogram of the breast mass diagnostic model. MIP: maximum intensity projection.

2.3 乳腺肿块诊断模型列线图的建立与评估

       采用多因素logistic回归分析并建立乳腺肿块诊断模型列线图(图3)。在诊断概率截断值取0.5时,乳腺肿块诊断模型的敏感度、特异度分别为95.90%、85.71%(训练集)和93.33%、83.93%(验证集)。DeLong检验显示在Kaiser评分中引入年龄、MIP征、附属影像特征构建的乳腺肿块诊断模型其区分度优于Kaiser评分(图4A4B),在训练集中,乳腺肿块诊断模型及Kaiser评分AUC分别为0.944、0.890,差异有统计学意义(P<0.05);NRI为0.1319,差异有统计学意义(Z=2.7272,P<0.05);IDI为0.1319,差异有统计学意义(Z=2.7111,P<0.05);在验证集中,乳腺肿块诊断模型模型及Kaiser评分的AUC分别为0.941、0.874,差异有统计学意义(P<0.05);NRI为0.2131,差异有统计学意义(Z=2.1697,P<0.05);IDI为0.2131,差异有统计学意义(Z=2.1348,P<0.05)。DCA表明在训练集及验证集中,在大范围阈值概率内,使用乳腺肿块诊断模型诊断预测乳腺肿块的净收益大于Kaiser评分(图4C-4D);校准曲线(图4E-4F)显示乳腺肿块诊断模型对病灶恶性概率预测值和实际结果之间有较好的一致性。模型具体应用举例见图5

图4  训练集及验证集中乳腺肿块诊断模型和Kaiser评分的诊断性能对比。4A~4D:训练集及验证集中乳腺肿块诊断模型和Kaiser评分的ROC曲线(Bootstrap=500次)、DCA曲线;4E~4F:乳腺肿块诊断模型分别在训练集及验证集中的校准曲线。ROC:受试者工作特征;DCA:决策曲线分析。
Fig. 4  Comparison of the diagnostic performance of the breast mass diagnostic model and Kaiser score in the training and validation sets. 4A-4D: ROC curves (Bootstrap=500 times), DCA curves of the breast mass diagnostic model and Kaiser score in the training and validation sets; 4E-4F: Calibration curves of the breast mass diagnostic model in the training and validation sets, respectively. ROC: receiver operating characteristic; DCA: ecision curve analysis.
图5  女,80岁,发现左侧乳腺肿块半月,稍疼痛,伴局部红肿及乳头溢液。5A:轴位T2-FS;5B:轴位T1WI;5C:轴位增强预扫描序列;5D:轴位增强扫描早期时相;5E~5F:病灶时间-信号曲线;5G:轴位MIP图;5H:增强扫描早期时相(示附属影像特征:左侧腋窝淋巴结大小及形态异常);5I:左侧乳腺肿块穿刺病理(HE,10×10);5J:左侧腋窝淋巴结穿刺病理(HE,10×4);本例表现为左侧乳腺乳头后方卵圆形肿块(5A~5E白色箭),边界清楚,增强扫描早期时相缓慢强化,内部呈环形强化,延迟时相时间-信号曲线呈平台型,根据Kaiser评分模型,本例病灶被评为2分,考虑为良性;然而,病灶的附属影像特征:邻近乳晕皮肤增厚(5A~5D橙色箭)、左侧腋窝异常淋巴结(5H白色箭)、最大信号强度投影(MIP)征阳性(5G白色箭)及患者年龄同时也是对诊断非常有价值的信息,考虑到这些信息的乳腺肿块诊断模型对该病灶的评分为123.3分,恶性概率>0.9;本例经术后穿刺病理证实为浸润性导管癌Ⅱ级,伴左侧腋窝淋巴结转移。
Fig. 5  A female, 80 years old, found a left breast mass for half a month, slightly painful, with localized erythema and nipple overflow. 5A: Axial T2-FS; 5B: Axial T1WI; 5C: Axial enhanced pre-scan sequence; 5D: Early time phase of axial enhanced scan; 5E-5F: Time-signal intensity curve of the lesion; 5G: Axial MIP map; 5H: Early time phase of enhanced scan (showing accessory imaging features: left axillary lymph node size and morphology abnormal); 5I: Left breast mass puncture pathology (HE, 10×10); 5J: Left axillary lymph node puncture pathology (HE, 10×4). This case shows an ovoid mass behind the left nipple (5A-5E white arrow) with clear borders, slow enhancement in the early time phase of the enhancement scan, circular enhancement in the interior, delayed time phase time-signal intensity curve in plateau type, according to the Kaiser scoring model, the lesion in this case was rated as 2 and considered benign. However, the accessory imaging features of the lesion: thickening of the adjacent areolar skin (5A-5D orange arrows), abnormal lymph nodes in the left axilla (5H white arrow), positive maximum intensity projection (MIP) sign (5G white arrow), and the patient's age were also valuable information for the diagnosis, and the diagnostic model of the breast mass taking into account this information scored the lesion 123.3 with a probability of malignancy>0.9. This case was confirmed by postoperative puncture pathology as invasive ductal carcinoma grade Ⅱ with left axillary lymph node metastasis.

3 讨论

       本研究基于Kaiser评分,探讨了对乳腺肿块具有诊断价值的其他参数,并采用多因素logistic回归的方法在Kaiser评分中引入新的诊断参数,构建了一个乳腺肿块诊断模型列线图。在针对Kaiser评分的优化问题上,首次将“年龄”作为新的参数引入评分系统,并首次将Kaiser评分构建过程中已排除的诊断参数“附属影像特征”和“MIP征”重新引入评分系统,模型在区分度、校准度和临床有效性三个维度整体表现满意。在乳腺肿块的MRI诊断中,乳腺肿块诊断模型相对于Kaiser评分,其对病灶良恶性的区分度更好,临床有效性更高。

3.1 将“年龄”作为新的诊断参数引入Kaiser评分的临床合理性分析

       既往针对Kaiser评分优化问题的相关研究中,新引入参数基本集中在影像类的参数如合成磁共振成像(synthetic magnetic resonance imaging, SyMRI)中的T1值[13]和表观扩散系数[19, 20, 21],考虑到临床资料对影像诊断的参考意义,本研究首次在影像诊断参数之外探讨了对乳腺肿块有诊断价值的临床类参数,发现年龄对乳腺肿块具有诊断价值并引入了模型。年龄是乳腺癌发病主要影响因素之一,一项针对1990至2019年中国女性乳腺癌发病率与死亡率情况的相关研究报道表明,在校正时期效应与队列效应后,我国女性乳腺癌发病率的纵向年龄曲线呈单调上升趋势[22],另一些研究则表明,其原因可能是年龄相关性DNA甲基化,与年龄相关的DNA甲基化可能随时间推移而促进癌症的发生[17,23, 24]。这从一定程度上解释了在本研究中,年龄被纳入到乳腺疾病的诊断模型中并提升了Kaiser评分的诊断效能的原因。

3.2 将“MIP征”和“附属影像特征”重新引入Kaiser评分的临床合理性分析

       在Kaiser评分最初构建的过程中,从包括“MIP征”和“附属影像特征”在内的17个参数中提取出了最终纳入Kaiser评分的5个参数(“毛刺征”“TIC”“边缘”“内部强化模式”“水肿”),其参数的提取使用的方法是卡方自动交互检测(CHAID)[11],本研究尝试了用新的变量筛选方法(LASSO回归)和参数引入方法(多因素logistic回归)将被排除的“MIP征”和“附属影像特征”重新纳入基于Kaiser评分的诊断模型中,在本研究数据集中取得了改善Kaiser评分的诊断效能的结果。推测可能原因如下:(1)Kaiser评分出于评分结构树精简的考虑,纳入的是17个诊断参数中5个最有诊断价值的参数,其他被排除的诊断参数虽未入选但仍然具有一定的诊断价值;(2)不同研究对象变量分布的差异和变量筛选方法的不同导致了在不同数据集中变量筛选结果存在一定差异。本研究所构建乳腺肿块诊断模型基于中国人群数据,相较于Kaiser评分,可能更适合在中国人群中推广使用。

3.3 本研究的临床价值分析

       对经典Kaiser评分进行改良以提升其对乳腺良恶性病变的鉴别能力是近年来相关领域同行学者比较关心的问题,在引入新诊断参数方面,相关研究主要关注引入表观扩散系数(apparent diffusion coefficient, ADC)值可否提升Kaiser评分的诊断性能,且研究结果基本一致:组合使用ADC值和Kaiser评分与单独使用Kaiser评分相比,特异度轻度增加,但同时敏感度有所降低,整体诊断性能提升并不明显[19,25]。另外,SyMRI是最近提出的一种多动态多回波序列[26],近期有相关研究发现,将SyMRI中的T1值与Kaiser评分结合起来,对Kaiser评分鉴别乳腺良恶性病变具有附加价值[13],但SyMRI的实现需要额外的采集序列和分析软件,其应用在国内尚不普遍,截至目前,关于使用SyMRI评估乳腺病变的相关研究报道相对较少,并且结果并不一致[27, 28, 29],其在乳腺病变中的诊断应用尚需通过更多临床研究进行验证。在引入参数的方法方面,相关研究大多采用了将新引入的诊断参数和Kaiser评分值按截断值作为两个分层节点组合成树状结构的方式,这种方法简单、易操作,但新引入诊断参数和Kaiser评分作为评分策略中的不同组成部分,其权重分配较为主观。

       本研究所开发的乳腺肿块诊断模型新纳入的三个诊断参数“年龄”“MIP征”“附属影像特征”皆为临床实践中非常易得的信息,在提升Kaiser评分诊断性能的同时,不会增加额外的检查成本,适合在临床实际工作中进行推广;并且在引入新参数的方法方面,采用了多因素logistic回归,使各参数在诊断策略中的权重分配更细致和更具客观性。

3.4 本研究的局限性

       本研究存在一定的局限性。第一,本研究针对的是乳腺MRI肿块的诊断问题,模型不能适用于乳腺MRI非肿块类病变的诊断,在后续研究中,可以本研究构建的模型为基础,开发普遍适用于乳腺肿块和非肿块类病变的MRI诊断预测模型,使模型的临床应用范围更广;第二,本研究构建的乳腺肿块诊断预测模型虽然进行了外部验证,但验证集仅来源于一个中心且样本量相对有限,有待于在后续研究中扩大参与机构数和样本量对模型进行进一步的验证和校准。

4 结论

       本研究以Kaiser评分为基础,通过采用多因素logistic回归集成其他对乳腺肿块有诊断价值的参数的方法,构建了一个乳腺肿块诊断模型,经过验证,该诊断模型可以用于乳腺肿块恶性概率的术前预测,并且其对乳腺肿块的诊断性能优于经典Kaiser评分。

[1]
LEI S Y, ZHENG R S, ZHANG S W, et al. Global patterns of breast cancer incidence and mortality: a population-based cancer registry data analysis from 2000 to 2020[J]. Cancer Commun (Lond), 2021, 41(11): 1183-1194. DOI: 10.1002/cac2.12207">10.1002/cac2.12207">10.1002/cac2.12207.
[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]. CA Cancer J Clin, 2021, 71(3): 209-249. DOI: 10.3322/caac.21660">10.3322/caac.21660">10.3322/caac.21660.
[3]
LI J, CHEN C, NIE J J, et al. Changes in the disease burden of breast cancer along with attributable risk factors in China from 1990 to 2019 and its projections: an analysis of the global burden of disease study 2019[J]. Cancer Med, 2023, 12(2): 1888-1902. DOI: 10.1002/cam4.5006">10.1002/cam4.5006">10.1002/cam4.5006.
[4]
于洗河, 张景茹, 降海蕊, 等. 中国女性1990—2019年宫颈癌和乳腺癌疾病负担分析[J]. 中国公共卫生, 2022, 38(5): 534-538. DOI: 10.11847/zgggws1136776">10.11847/zgggws1136776">10.11847/zgggws1136776.
YU X H, ZHANG J R, JIANG H R, et al. Disease burden of cervical cancer and breast cancer in Chinese women in 1990 and 2019[J]. Chin J Public Health, 2022, 38(5): 534-538. DOI: 10.11847/zgggws1136776">10.11847/zgggws1136776">10.11847/zgggws1136776.
[5]
ARNOLD M, MORGAN E, RUMGAY H, et al. Current and future burden of breast cancer: global statistics for 2020 and 2040[J]. Breast, 2022, 66: 15-23. DOI: 10.1016/j.breast.2022.08.010">10.1016/j.breast.2022.08.010">10.1016/j.breast.2022.08.010.
[6]
WILKINSON L, GATHANI T. Understanding breast cancer as a global health concern[J/OL]. Br J Radiol, 2022, 95(1130): 20211033 [2022-12-07]. https://pubmed.ncbi.nlm.nih.gov/34905391/. DOI: 10.1259/bjr.20211033">10.1259/bjr.20211033">10.1259/bjr.20211033.
[7]
PARK A R, CHAE E Y, CHA J H, et al. Preoperative breast MRI in women 35 years of age and younger with breast cancer: benefits in surgical outcomes by using propensity score analysis[J]. Radiology, 2021, 300(1): 39-45. DOI: 10.1148/radiol.2021204124">10.1148/radiol.2021204124">10.1148/radiol.2021204124.
[8]
MANN R M, CHO N, MOY L. Breast MRI: state of the art[J]. Radiology, 2019, 292(3): 520-536. DOI: 10.1148/radiol.2019182947">10.1148/radiol.2019182947">10.1148/radiol.2019182947.
[9]
COMSTOCK C E, GATSONIS C, NEWSTEAD G M, 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): 746-756. DOI: 10.1001/jama.2020.0572">10.1001/jama.2020.0572">10.1001/jama.2020.0572.
[10]
WOITEK R, SPICK C, SCHERNTHANER M, et al. A simple classification system (the Tree flowchart) for breast MRI can reduce the number of unnecessary biopsies in MRI-only lesions[J]. Eur Radiol, 2017, 27(9): 3799-3809. DOI: 10.1007/s00330-017-4755-6">10.1007/s00330-017-4755-6">10.1007/s00330-017-4755-6.
[11]
BALTZER P A, DIETZEL M, KAISER W A. A simple and robust classification tree for differentiation between benign and malignant lesions in MR-mammography[J]. Eur Radiol, 2013, 23(8): 2051-2060. DOI: 10.1007/s00330-013-2804-3">10.1007/s00330-013-2804-3">10.1007/s00330-013-2804-3.
[12]
安永玉, 刘畅, 张宏霞, 等. Kaiser评分对MRI乳腺影像报告与数据系统4类病灶的诊断价值[J]. 浙江医学, 2021, 43(5): 511-515. DOI: 10.12056/j.issn.1006-2785.2021.43.5.2020-3139">10.12056/j.issn.1006-2785.2021.43.5.2020-3139">10.12056/j.issn.1006-2785.2021.43.5.2020-3139.
AN Y Y, LIU C, ZHANG H X, et al. Diagnostic value of Kaiser score for BI-RADS 4 lesions on MRI[J]. Zhejiang Med J, 2021, 43(5): 511-515. DOI: 10.12056/j.issn.1006-2785.2021.43.5.2020-3139">10.12056/j.issn.1006-2785.2021.43.5.2020-3139">10.12056/j.issn.1006-2785.2021.43.5.2020-3139.
[13]
MENG L S, ZHAO X, GUO J X, et al. Evaluation of the differentiation of benign and malignant breast lesions using synthetic relaxometry and the Kaiser score[J/OL]. Front Oncol, 2022, 12: 964078 [2022-12-07]. https://pubmed.ncbi.nlm.nih.gov/36303839/. DOI: 10.3389/fonc.2022.964078">10.3389/fonc.2022.964078">10.3389/fonc.2022.964078.
[14]
ISTOMIN A, MASARWAH A, VANNINEN R, et al. Diagnostic performance of the Kaiser score for characterizing lesions on breast MRI with comparison to a multiparametric classification system[J/OL]. Eur J Radiol, 2021, 138: 109659 [2022-12-07]. https://pubmed.ncbi.nlm.nih.gov/33752000/. DOI: 10.1016/j.ejrad.2021.109659">10.1016/j.ejrad.2021.109659">10.1016/j.ejrad.2021.109659.
[15]
周海鹰, 陈天武, 张小明. 乳腺血供不对称性增加及邻近血管征鉴别诊断乳腺良恶性病变[J]. 中国医学影像技术, 2016, 32(6): 900-904. DOI: 10.13929/j.1003-3289.2016.06.021">10.13929/j.1003-3289.2016.06.021">10.13929/j.1003-3289.2016.06.021
ZHOU H Y, CHEN T W, ZHANG X M. Asymmetric increase in breast vascularity and adjacent vessel sign in differential diagnosis of benign and malignant breast lesions[J]. Chin J Med Imaging Technol, 2016, 32(6): 900-904. DOI: 10.13929/j.1003-3289.2016.06.021">10.13929/j.1003-3289.2016.06.021">10.13929/j.1003-3289.2016.06.021
[16]
ÇETINKAYA E, YILDIZ Ş, OTÇU H, et al. The value of adjacent vessel sign in malignant breast tumors[J]. Diagn Interv Radiol, 2022, 28(5): 463-469. DOI: 10.5152/dir.2022.211228">10.5152/dir.2022.211228">10.5152/dir.2022.211228.
[17]
KRESOVICH J K, XU Z L, O'BRIEN K M, et al. Methylation-based biological age and breast cancer risk[J]. J Natl Cancer Inst, 2019, 111(10): 1051-1058. DOI: 10.1093/jnci/djz020">10.1093/jnci/djz020">10.1093/jnci/djz020.
[18]
BALTZER P A T, KRUG K B, DIETZEL M. Evidence-based and structured diagnosis in breast MRI using the Kaiser score[J]. Rofo, 2022, 194(11): 1216-1228. DOI: 10.1055/a-1829-5985">10.1055/a-1829-5985">10.1055/a-1829-5985.
[19]
潘佳玲, 李晓虹, 陈新杰, 等. Kaiser评分与表观扩散系数评价乳腺病灶的对比研究[J]. 磁共振成像, 2022, 13(6): 108-111, 116. DOI: 10.12015/issn.1674-8034.2022.06.021">10.12015/issn.1674-8034.2022.06.021">10.12015/issn.1674-8034.2022.06.021
PAN J L, LI X H, CHEN X J, et al. A comparison of the Kaiser score and apparent diffusion coefficient mapping in the assessment of breast lesions[J]. Chin J Magn Reson Imaging, 2022, 13(6): 108-111, 116. DOI: 10.12015/issn.1674-8034.2022.06.021">10.12015/issn.1674-8034.2022.06.021">10.12015/issn.1674-8034.2022.06.021
[20]
MENG L S, ZHAO X, LU L, et al. A comparative assessment of MR BI-RADS 4 breast lesions with Kaiser score and apparent diffusion coefficient value[J/OL]. Front Oncol, 2021, 11: 779642 [2022-12-07]. https://pubmed.ncbi.nlm.nih.gov/34926290/. DOI: 10.3389/fonc.2021.779642">10.3389/fonc.2021.779642">10.3389/fonc.2021.779642.
[21]
DIETZEL M, KRUG B, CLAUSER P, et al. A multicentric comparison of apparent diffusion coefficient mapping and the Kaiser score in the assessment of breast lesions[J]. Invest Radiol, 2021, 56(5): 274-282. DOI: 10.1097/RLI.0000000000000739">10.1097/RLI.0000000000000739">10.1097/RLI.0000000000000739.
[22]
刘雪薇, 王媛, 韦丹梅, 等. 1990—2019年中国女性乳腺癌发病及死亡趋势的年龄-时期-队列模型分析[J]. 中国全科医学, 2023(1): 34-41.
LIU X W, WANG Y, WEI D M, et al. Age-period-cohort analysis of trends of breast cancer incidence and mortality among Chinese females from 1990 to 2019[J]. Chin Gen Pract, 2023(1): 34-41.
[23]
KIELY M, TSE L A, KOKA H, et al. Age-related DNA methylation in paired normal and tumour breast tissue in Chinese breast cancer patients[J]. Epigenetics, 2021, 16(6): 677-691. DOI: 10.1080/15592294.2020.1819661">10.1080/15592294.2020.1819661">10.1080/15592294.2020.1819661.
[24]
MARINO N, GERMAN R, PODICHETI R, et al. Aberrant epigenetic and transcriptional events associated with breast cancer risk[J/OL]. Clin Epigenetics, 2022, 14(1): 21 [2022-12-07]. https://pubmed.ncbi.nlm.nih.gov/35139887/. DOI: 10.1186/s13148-022-01239-1">10.1186/s13148-022-01239-1">10.1186/s13148-022-01239-1.
[25]
CHEN Z W, ZHAO Y F, LIU H R, et al. Assessment of breast lesions by the Kaiser score for differential diagnosis on MRI: the added value of ADC and machine learning modeling[J]. Eur Radiol, 2022, 32(10): 6608-6618. DOI: 10.1007/s00330-022-08899-w">10.1007/s00330-022-08899-w">10.1007/s00330-022-08899-w.
[26]
ANDICA C, HAGIWARA A, HORI M, et al. Review of synthetic MRI in pediatric brains: basic principle of MR quantification, its features, clinical applications, and limitations[J]. J De Neuroradiol, 2019, 46(4): 268-275. DOI: 10.1016/j.neurad.2019.02.005">10.1016/j.neurad.2019.02.005">10.1016/j.neurad.2019.02.005.
[27]
MATSUDA M, TSUDA T, EBIHARA R, et al. Enhanced masses on contrast-enhanced breast: differentiation using a combination of dynamic contrast-enhanced MRI and quantitative evaluation with synthetic MRI[J]. J Magn Reson Imaging, 2021, 53(2): 381-391. DOI: 10.1002/jmri.27362">10.1002/jmri.27362">10.1002/jmri.27362.
[28]
MENG T B, HE N, HE H Q, et al. The diagnostic performance of quantitative mapping in breast cancer patients: a preliminary study using synthetic MRI[J/OL]. Cancer Imaging, 2020, 20(1): 88 [2022-12-07]. https://pubmed.ncbi.nlm.nih.gov/33317609/. DOI: 10.1186/s40644-020-00365-4">10.1186/s40644-020-00365-4">10.1186/s40644-020-00365-4.
[29]
GAO W B, ZHANG S Q, GUO J X, et al. Investigation of synthetic relaxometry and diffusion measures in the differentiation of benign and malignant breast lesions as compared to BI-RADS[J]. J Magn Reson Imaging, 2021, 53(4): 1118-1127. DOI: 10.1002/jmri.27435">10.1002/jmri.27435">10.1002/jmri.27435.

上一篇 动脉自旋标记成像诊断干燥综合征早期腮腺损伤的临床价值
下一篇 多参数MRI在原发性乳腺癌不同分子亚型中的诊断价值研究
  
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