分享:
分享到微信朋友圈
X
临床研究
乳腺X线及MRI特征联合临床病理预测乳腺导管原位癌伴微浸润
周晓平 杨蔚 尹清云 张宁妹 张朝林 刘开惠 吴林桦

Cite this article as ZHOU X P, YANG W, YIN Q Y, et al. Combining the X-ray and MRI characteristics with the clinical pathology to predict ductal carcinoma in situ with microinvasion of breast[J]. Chin J Magn Reson Imaging, 2024, 15(5): 102-110, 118.本文引用格式周晓平, 杨蔚, 尹清云, 等. 乳腺X线及MRI特征联合临床病理预测乳腺导管原位癌伴微浸润[J]. 磁共振成像, 2024, 15(5): 102-110, 118. DOI:10.12015/issn.1674-8034.2024.05.017.


[摘要] 目的 探讨基于临床病理、乳腺X线(mammography, MG)和MRI特征预测乳腺导管原位癌伴微浸润(ductal carcinoma in situ with microinvasion, DCISM)的价值。材料与方法 回顾性收集宁夏医科大学总医院2019年6月至2022年6月最终经手术病理证实为纯导管原位癌(ductal carcinoma in situ, DCIS)和DCISM的首诊女性患者的病例资料为训练组,评估术前患者的临床病理、MG和MRI特征。采用单、多因素logistic回归分析明确DCISM的独立危险因素,并建立联合模型。通过受试者工作特征(receiver operating characteristic, ROC)曲线下面积(area under the curve, AUC)及校准图评估模型的诊断效能,应用决策曲线分析(decision curve analysis, DCA)观察联合模型的临床实用性。前瞻性收集2022年7月至2023年7月符合纳入、排除标准的患者作为验证组进行验证。应用沙普利加和解释(Shapley Additive exPlanation, SHAP)分析联合模型中病灶最长径、核级别、坏死、Ki-67指数、P63状态、钙化状态和最小ADC值(minimum value of apparent diffusion coefficient, ADCmin)预测DCISM的价值。收集535例患者共550个病灶(15例患者为同时性双乳癌),患者年龄23~81岁,中位年龄50岁。训练组(n=382)中102个病灶(27%)和验证组(n=168)中52个病灶(31%)被诊断为DCISM。结果 多因素logistic回归结果显示,病灶最长径、核级别、坏死、Ki-67指数、P63状态、钙化状态和ADCmin是DCISM的独立危险因素。基于上述参数构建临床病理及联合模型,在训练组和验证组中均表现出较高的预测效能(AUC:0.937、0.899)。根据SHAP分析,病灶最长径、Ki-67指数和ADCmin在联合模型中对预测DCISM起主要贡献,而钙化状态、核级别、P63状态和坏死是补充因素。结论 联合临床病理及术前MG和MRI特征的预测模型可有效从纯DCIS区分出DCISM,从而提升临床决策和治疗规划的准确性。
[Abstract] Objective To explore the value of clinical-pathological, mammographic (MG), and MRI features in predicting ductal carcinoma in situ with microinvasion (DCISM).Materials and Methods A retrospective study was conducted on female patients diagnosed with pure ductal carcinoma in situ (DCIS) and DCISM confirmed by final surgical pathology from June 2019 to June 2022 at General Hospital of Ningxia Medical University. Clinical-pathological, MG, and MRI features of the patients were evaluated. The univariate and multivariate logistic regression analysis was used to identify independent risk factors for DCISM and develop a combined model. The diagnostic performance of the model was assessed using the area under the receiver operating characteristic (ROC) curve (AUC) and calibration plot. The clinical utility of the combined model was evaluated using decision curve analysis (DCA). A prospective validation was performed on patients who meet the eligibility criteria for inclusion and exclusion from July 2022 to July 2023. Shapley Additive exPlanation (SHAP) analysis was applied to assess the value of the combined model in predicting DCISM based on the longest diameter of the lesion, nuclear grade, necrosis, Ki-67 index, P63 status, calcification status, and minimum ADC value. A total of 535 patients with 550 lesions (15 cases were synchronous bilateral breast cancer) were collected. The patients' ages ranged from 23 to 81 years, with a median age of 50 years. Among the training group (n=382), 102 lesions (27%) were diagnosed as DCISM, while in the validation group (n=168), 52 lesions (31%) were diagnosed as DCISM.Results The multivariable logistic regression analysis showed the independent risk factors of DCISM included longest diameter of the lesion, nuclear grade, necrosis, Ki-67 index, P63 status, calcification status, and the minimum value of apparent diffusion coefficient (ADCmin). A predictive model combining the above parameters with preoperative clinical-pathological, mammography, and MRI features was constructed, demonstrating high predictive performance in both the training and validation groups (AUC: 0.937, 0.899). According to SHAP analysis, the longest diameter of the lesion, Ki-67 index, and ADCmin make the primary contributions in the combined model for predicting DCISM, while the calcification status, nuclear grade, P63 status, and necrosis are supplementary factors.Conclusions A combined predictive model using clinical-pathological, preoperative MG and MRI features can effectively differentiate DCISM from pure DCIS, thereby improving the accuracy of clinical decision-making and treatment planning.
[关键词] 乳腺肿瘤;导管原位癌;导管原位癌伴微浸润;可解释性;乳腺X线摄影;磁共振成像
[Keywords] breast tumor;ductal carcinoma in situ;ductal carcinoma in situ with microinvasion;interpretability;mammography;magnetic resonance imaging

周晓平 1   杨蔚 2*   尹清云 3   张宁妹 4   张朝林 5   刘开惠 1   吴林桦 2  

1 宁夏医科大学临床医学院,银川 750004

2 宁夏医科大学总医院放射科,银川 750004

3 宁夏医科大学总医院肿瘤内科,银川 750004

4 宁夏医科大学总医院病理科,银川 750004

5 宁夏医科大学总医院肿瘤外科,银川 750004

通信作者:杨蔚,E-mail:yangwei_0521@163.com

作者贡献声明::杨蔚设计本研究的方案,对稿件重要内容进行了修改,获得2022年宁夏回族自治区重点研发计划项目资助;周晓平起草和撰写稿件、获取、分析、解释本研究的数据;尹清云、张宁妹、张朝林、刘开惠、吴林桦获取、分析或解释本研究的数据,对稿件重要内容进行了修改;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: 2022年宁夏回族自治区重点研发计划项目 2022BEG03166
收稿日期:2023-08-15
接受日期:2024-04-17
中图分类号:R445.2  R737.9 
文献标识码:A
DOI: 10.12015/issn.1674-8034.2024.05.017
本文引用格式周晓平, 杨蔚, 尹清云, 等. 乳腺X线及MRI特征联合临床病理预测乳腺导管原位癌伴微浸润[J]. 磁共振成像, 2024, 15(5): 102-110, 118. DOI:10.12015/issn.1674-8034.2024.05.017.

0 引言

       乳腺导管原位癌(ductal carcinoma in situ, DCIS)是一种早期的非浸润性乳腺癌的病理阶段,尚未侵犯乳腺导管以外的区域[1]。伴随乳腺X线(mammography, MG)筛查的广泛推荐应用,DCIS的发现率明显增高,约占新发乳腺癌病例的20%~30%[2]。DCIS伴微浸润(ductal carcinoma in situ with microinvasion, DCISM)主要由DCIS和少量浸润性癌灶(侵袭深度≤1 mm)构成[3]。DCIS伴侵犯的标准治疗[4]是手术切除,并根据复发转移的危险度,而选择同时性前哨淋巴结活检(sentinel lymph node biopsy, SLNB)或腋窝淋巴结清扫(axillary lymph node dissection, ALND)。对于DCISM患者,需同时行SLNB以评估淋巴结状况。研究表明由于穿刺活检的局限性,导致约59% DCIS患者术后升级为浸润性癌[5],因此部分外科医生会将术前穿刺活检证实为DCIS的患者进行同时性SLNB,甚至进行ALND以防二次手术。然而,部分纯DCIS不会威胁患者生命,甚至终生不会进展为浸润性癌[6],因此不需要腋窝淋巴结评估,甚至可以密切随诊观察,而无需手术,也就避免了腋窝淋巴管水肿[7]及感觉异常[8]等并发症的出现。因此,术前准确评估DCISM可帮助医生选择最佳的治疗策略,对临床具有重要意义。

       动态对比增强MRI(dynamic contrast-enhanced MRI, DCE-MRI)对DCIS和浸润性癌的敏感性很高[9, 10]。在DCE-MRI扫描中,较大的病灶直径、肿块型病灶和不均匀强化被认为是DCIS发生浸润性癌的危险因素[11, 12, 13]。此外,经扩散加权成像(diffusion-weighed imaging, DWI)得到的表观扩散系数(apparent diffusion coefficient, ADC)值在有浸润性癌成分的患者中比DCIS患者低[5]。然而当前多数研究仅侧重于DCIS升级为较大浸润性病灶(通常浸润性成分的最大径线超过1.0 cm)的研究[14],较少单独研究微小浸润性病变,目前未见基于临床病理及术前MG和MRI特征的联合预测模型。DCISM作为隐匿的浸润性癌灶,对于放射科医师来说,从DCIS中预测DCISM一直是研究的重点、难点。

       因此,本研究旨在构建基于临床病理与术前MG和MRI特征的联合模型预测DCISM,并应用沙普利加和解释(Shapley Additive exPlanation, SHAP)[15, 16, 17]分析明确模型中每个变量的贡献价值,为准确识别DCIS中隐匿性微侵袭性癌灶提供可靠方法,以提升临床决策和治疗规划的准确性。

1 材料与方法

1.1 临床资料

       本研究遵守《赫尔辛基宣言》,已被宁夏医科大学总医院研究伦理委员会批准,免除训练组受试者知情同意,伦理批号:KYLL-2022-0251。所有验证组受试者均签署知情同意书。回顾性收集宁夏医科大学总医院手术病理证实为纯DCIS和DCISM的首诊女性患者的病例资料作为训练组。前瞻性收集2022年7月至2023年7月符合纳入、排除标准的患者作为验证组。训练组和验证组采用相同的MG和MRI检查方案。训练组纳入标准:(1)首诊患者,且最终经手术病理证实为DCIS或DCISM;(2)在术前、经皮穿刺或微创旋切(vacuum-assisted breast biopsy, VABB)前进行MG和MRI检查;(3)MG和MRI检查时间间隔不超过45天;(4)患乳单病灶。排除标准:(1)既往有乳腺手术史,比如因乳腺纤维瘤、乳头状瘤、乳腺炎等良性病灶切除以及乳腺癌保乳手术等;(2)病理结果不完整;(3)图像质量欠佳。验证组纳入标准:(1)年龄与训练组相匹配的初诊女性患者;(2)临床可疑DCIS/DCISM或经皮穿刺、VABB后初步诊断为DCIS/DCISM。排除标准:(1)既往有乳腺手术史;(2)MG和MRI检查间隔超过45天;(3)存在MRI检查禁忌;(4)影像图像质量差,无法用于诊断。

1.2 影像检查方法

       所有患者均应接受美国GE Senographe DS全数字化乳腺X线机和美国GE Signa HDxt 1.5 T乳腺MRI仪(包括DCE-MRI和DWI)检查。嘱患者立位,应用乳腺X线机检查,常规用全自动控制曝光,28~65 mAs,25~33 kV,摄取双乳头尾(cranio-caudal, CC)位和内外斜(mediolateral oblique , MLO)位像,当CC位和MLO位病灶未见显示,而临床触诊阳性或CC位和MLO位仅有一个体位观察到病灶,需进一步了解病灶的结构时需加摄90°方位像。乳腺MRI检查要求患者俯卧位于检查床,头先进,双乳自然悬垂于专用乳腺8通道线圈中。扫描序列与参数如下:轴位DWI检查(b=0和1000 s/mm2),TR 8000 ms,TE 65 ms,层厚4 mm,层间距1 mm,FOV 320 mm×320 mm,矩阵132×132,激励次数(number of excitation, NEX)4,扫描时间85 s;横断位DCE-MRI梯度回波序列脂肪抑制T1加权,翻转角15°,TR 4.7 ms,TE 2.3 ms,层厚 2 mm,层间距0 mm,FOV 330 mm×330 mm,矩阵416×416,NEX 0.8,扫描时间432 s。DCE-MRI检查时通过高压注射器(Bayer AG, MEDRAD Spectris Solaris EP)经肘前静脉以2.0 mL/s快速团注钆喷酸葡胺注射液(Gd-DTPA,商品名马根维显,拜耳医药保健股份有限公司生产,0.2 mmol/kg),并使用20 mL 0.9% NaCl溶液作为冲洗液。在注射对比剂前行蒙片扫描,前后连续扫描8个时相,每个时相的采集时间为54 s。

1.3 图像分析

       影像分析由两名放射科医生(分别具有22年和3年乳腺影像诊断经验的主任医师和住院医师)在对临床病理结果不知晓的情况下独立阅片。根据第5版美国放射学学会乳腺影像报告和数据系统[18]的指南,对MG和MRI图像解读。使用GE ADW 4.4工作站FUNCTION软件将导入的原始图像进行后处理。参照T2WI,在DWI上选择包含肿瘤实体成分的连续层面,沿肿瘤边缘手工勾画感兴趣区(region of interest, ROI),从对应的ADC伪彩图中,记录所有选定层面的ADC值,其平均值作为肿瘤平均ADC(ADCmean)[19]。含肿瘤实性成分的各个层面放置20~30个圆形ROI(40~50 mm2),记录每一个ROI对应的ADC值,将最小ADC和最大ADC值分别作为肿瘤ADCmin和ADCmax。选取ROI避开肿瘤坏死、囊性或出血部位。应用公式(1)[20]计算肿瘤异质性ADC值。选择DCE-MRI的晚期时相,肿瘤实体成分的最大层面上手动放置一个约40~50 mm2大小的圆形ROI,随后通过FUNCTION软件自动生成时间-信号曲线(time-intensity curve, TIC),包括流入型/Ⅰ型、平台型/Ⅱ型和流出型/Ⅲ型[21]。在轴位DCE-MRI扫描的晚期时相上测量病变的最长直径,作为DCIS/DCISM的大小。病变强化类型分为肿块样强化和非肿块样强化[22],内部强化方式包括均匀、不均匀、团块样及簇环样强化[23]

1.4 病理分析

       由一位具有24年乳腺病理诊断经验的副主任医师分析评估手术切除标本的病理类型、核级别、有无坏死、雌激素受体(estrogen receptor, ER)状态、孕激素受体(progesterone receptor, PR)状态、人表皮生长因子受体-2(human epidermal growth factor receptor 2, HER-2)状态、Ki-67指数和P63状态。通过免疫组织化学(immunohistochemistry, IHC)[24]检测到核染色比率为1%~100%的肿瘤细胞被定义为ER/PR阳性。HER-2阳性被定义为具有HER-2 IHC得分3+和/或通过荧光原位杂交(fluorescence in situ hybridization, FISH)检测到基因扩增的HER-2 IHC得分为2+。腋窝状态被定义为至少有一个经病理学确认的腋窝淋巴结存在。

1.5 联合模型的建立、诊断效能评估和验证

       以DCIS是否伴有微浸润作为因变量,进行单、多因素logistic 回归分析(将单因素分析后P≤0.10的参数纳入多因素分析),最终筛选出独立危险因素(P<0.05),建立联合预测模型。绘制受试者工作特征(receiver operating characteristic, ROC)曲线评估诊断效能。使用决策曲线分析(decision curve analysis, DCA)和校准曲线分别评估联合模型的临床净收益和预测结果与实际结果之间的一致性。使用Hosmer-Lemeshow检验评估联合模型的拟合优度。

1.6 联合模型的可解释性

       SHAP分析可公正评估每个变量对所构建模型诊断效能的贡献度[16]。利用R软件(https://cran.r-project.org)计算各个变量SHAP值大小,并由高到低对变量进行重要性排序,绘制经典直方图。SHAP值越大,表明该变量对模型的贡献越大。反之,贡献越小。

1.7 统计学分析

       使用SPSS 25.0软件(IBM Corp., Armonk, NY, USA)、MedCalc 20.217(MedCalc Software Ltd., Ostend, Belgium)和R 4.1.1(R Development Core Team, Vienna, Austria)软件进行分析。连续数据以均值±标准差或中位数(四分位数间距)表示,分类数据以例数或百分比表示。两组间特征比较采用独立样本t检验、Mann-Whitney U检验、卡方检验或Fisher精确检验。采用组内相关系数(intra-class correlation coefficient, ICC)和Kappa统计量分析观察者间一致性[25],0<ICC≤0.50为一致性差,0.50<ICC≤0.75为一致性中等,0.75<ICC≤0.90为一致性良好,0.90<ICC≤1.00为一致性好;0<Kappa≤0.20为一致性差,0.20<Kappa≤0.40为一致性较差,0.40<Kappa≤0.60为一致性中等,0.60<Kappa≤0.80为一致性良好,0.80<Kappa≤1.00为一致性好。当P<0.05时,认为差异有统计学意义。

2 结果

2.1 一般资料

       纳入535名患者(15名同时性双侧乳腺癌),共550个病灶。其中,训练组373名患者(9名同时性双侧乳腺癌),共382个病灶,有102个病灶(27%)被诊断为DCISM;验证组162名患者(6名同时性双侧乳腺癌),共168个病灶,有52个病灶(31%)被诊断为DCISM。患者年龄23~81岁,中位年龄50岁。

2.2 影像变量的观察者间一致性分析

       两名诊断医师之间的ICC值和Kappa值均≥0.75,表明两名观察者间的一致性水平中等至良好(表1)。

表1  MG和MRI变量的观察者间一致性
Tab. 1  Interobserver agreement for MG and MRI variables

2.3 训练组和验证组间基线特征比较

       训练组和验证组的临床、病理和影像参数比较,差异均无统计学意义(P均>0.05)。两组间具体基线特征见表2

表2  训练组和验证组间临床、病理和影像参数比较
Tab. 2  Comparison of clinical, pathological, and imaging parameters between the training group and the validation group

2.4 训练组中DCIS和DCISM组间特征比较

       训练组中DCIS和DCISM组间特征比较发现,病灶最长径、核级别、坏死率、ER状态、HER-2状态、Ki-67指数、P63状态、MG特征、钙化状态、病灶强化类型及内部强化方式、ADCmin以及ADC异质性在两组间差异有统计学意义(P均<0.05)(表3)。

表3  训练组纯DCIS和DCISM临床、病理和影像参数比较
Tab. 3  Comparison of clinical, pathological, and imaging parameters between pure DCIS and DCISM in the training group

2.5 训练组DCISM独立危险因素

       单因素和多因素logistic回归分析显示:病灶最长径、核级别、坏死率、Ki-67指数、P63状态、钙化状态以及ADCmin是DCIS升级的独立危险因素(表4)。

表4  与DCISM相关因素的单因素和多因素logistic回归分析
Tab. 4  Univariate and multivariate logistic regression analyses of factors associated with DCISM

2.6 联合模型的建立、诊断效能评估和验证

       基于上述参数建立联合模型(表5)。该模型在训练组中显示出较高的预测能力,ROC曲线下面积(area under the curve, AUC)为0.937(95% CI: 0.907~0.959),高于任何单个危险因素的AUC(图1A)。在训练组中,该模型经Hosmer-Lemeshow检验,预测结果与观察结果之间具有良好的一致性(P=0.125)(图1B)。DCA表明,联合模型对预测为DCISM的患者比不预测的患者获得更高的净收益(图1C)。在验证组中联合模型也表现出良好的诊断效能(AUC:0.899,95% CI:0.843~0.940,P<0.001)、校准能力(Hosmer-Lemeshow检验,P=0.228)和临床决策能力(图1D~1F)。

图1  评估联合模型在训练组和验证组中的诊断效能。1A:训练组中联合模型与各独立危险因素的ROC曲线;1B:训练组中联合模型的校准图;1C:训练组中联合模型的决策曲线图;1D:验证组中联合模型的ROC曲线;1E:验证组中联合模型的校准图;1F:验证组中联合模型的决策曲线图。1C、1F中蓝色虚线代表没有微浸润,绿色虚线代表包括所有有微浸润的情况,红色实线代表联合模型,表示在所有阈值概率下,联合模型在训练组与验证组中,人群的临床净收益均较高。ROC:受试者工作特征;Ki-67:增殖细胞核抗原;P63:肌上皮标记物;ADC:表观扩散系数;ADCmin:最小ADC值。
Fig. 1  Evaluation of the diagnostic performance of the combined model in the training group and validation group. 1A: ROC curves of the combined model and individual risk factors in the training group; 1B: Calibration plot of the combined model in the training group; 1C: Decision curve analysis plot of the combined model in the training group; 1D: ROC curve of the combined model in the validation group; 1E: Calibration plot of the combined model in the validation group; 1F: Decision curve analysis plot of the combined model in the validation group. In 1C and 1F, the blue dashed line represents no DCISM, the green dashed line represents all cases with DCISM, and the red solid line represents the combined model. It indicates that the combined model has higher clinical net benefit in the population of the training and validation groups across all threshold probabilities. ROC: receiver operating characteristic; DCISM: ductal carcinoma in situ with microinvasion; Ki-67: proliferation cell nuclear antigen; P63: a marker for myoepithelial cells; ADC: apparent diffusion coefficient; ADCmin: the minimum ADC value.
表5  DCISM的独立危险因素及联合模型的预测效能
Tab. 5  Predicting performance of DCISM independent risk factors and the combined model

2.7 联合模型的可解释性和临床获益

       通过SHAP分析,在联合模型中,病灶最长径、Ki-67指数和ADCmin对预测DCISM起着主要贡献,而钙化状态、核级别、P63状态和坏死是补充因素(图2),图3举例说明该联合模型的临床应用,即模型预测结果与最终病理结果的一致性。

图2  应用SHAP分析联合模型中各独立危险因素的贡献并降序排列。SHAP值越大,相应的参数贡献越大。SHAP:沙普利加和解释;Ki-67:增殖细胞核抗原;P63:肌上皮标记物;ADC:表观扩散系数;ADCmin:最小ADC值。
Fig. 2  Applying SHAP analysis to assess the contributions of individual risk factors in the combined model and ranking them in descending order. Larger SHAP values indicate greater contributions of the corresponding parameters. SHAP: Shapley Additive exPlanation; Ki-67: proliferation cell nuclear antigen; P63: a marker for myoepithelial cells; ADC: apparent diffusion coefficient; ADCmin: the minimum ADC value.
图3  女,46岁,右乳外象限肿物,临床可疑DCIS,穿刺活检后免疫组化显示该患者P63表达阴性,Ki-67低表达(15%)。3A:乳腺X线示右乳MLO位呈节段性、簇状分布的细小多形性可疑钙化(红色五边形);3B:横断位DWI示病灶呈高信号(白箭);3C:演示在相对应的ADC伪彩图上,尽可能多地放置圆形或椭圆形ROI,同时避开囊性、坏死或出血区域(白箭);3D:免疫组化图示Ki-67表达约15%(×400,蓝箭);3E:免疫组化图示P63阴性表达(×400,蓝色五角星);3F:该病例的SHAP图显示,病灶最长径和Ki-67对联合模型的贡献最大,蓝色和绿色区域分别表示参数对预测风险的正向和负向得分。当模型的预测值f(x)大于截断值E[f(x)]时,预示DCISM的可能性大。此病例模型的预测值f(x)为52.5,大于截断值E[f(x)]=50,故模型提示提示DCISM可能,该预测结果与患者的术后病理一致。DCIS:导管原位癌;P63:肌上皮标记物;Ki-67:增殖细胞核抗原;MLO位:内外侧斜位;DWI:扩散加权成像;ADC:表观扩散系数;ROI:感兴趣区域;SHAP:沙普利加和解释;DCISM:乳腺导管原位癌伴微浸润;ADCmin:最小ADC 值。
Fig. 3  Female, 46 years old, with a mass in the outer quadrant of the right breast, clinically suspicious for DCIS. After a puncture biopsy, immunohistochemistry revealed that the patient is negative for P63 expression and has low expression of Ki-67 (15%). 3A: Mammography shows segmental and clustered distribution of fine heterogeneous suspicious calcifications (red pentagons) in the MLO position of the right breast; 3B: Transverse DWI shows a lesion with high signal intensity (white arrow); 3C: Demonstrated on the corresponding pseudocolored ADC map, circular or elliptical ROIs are placed as much as possible, while avoiding cystic, necrotic, or hemorrhagic areas (white arrow); 3D: Immunohistochemical staining shows approximately 15% Ki-67 expression (×400 magnification, blue arrow); 3E: Immunohistochemical staining shows negative expression of P63 (×400 magnification, blue pentagram); 3F: The SHAP plot of this case reveals that the longest diameter of the lesion and Ki-67 have the highest contributions to the combined model. The blue and green regions represent positive and negative scores of parameters in predicting risk, respectively. When the model's predicted value f(x) exceeds the threshold value E[f(x)], it suggests a higher likelihood of DCISM. For this case, the predicted value f(x) of the model is 52.5, which is greater than the cutoff value E[f(x)]=50. Therefore, the model suggests a probability of DCISM, and this prediction is consistent with the patient's postoperative pathology. DCIS: ductal carcinoma in situ; P63: a marker for myoepithelial cells; Ki-67: proliferation cell nuclear antigen; MLO: mediolateral oblique; DWI: diffusion-weighed imaging; ADC: apparent diffusion coefficient; ROI: region of interest; SHAP: Shapley Additive exPlanation; DCISM: ductal carcinoma in situ with microinvasion; ADCmin: the minimum ADC value.

3 讨论

       本研究基于患者术前MG和MRI影像特征及临床病理,建立联合模型,并首次利用SHAP分析明确各独立危险因素在联合模型中的贡献大小,从而探讨该模型在临床可疑DCIS、经空心针穿刺活检(core needle biopsy, CNB)或术中病理证实为DCIS的首诊患者中准确识别出是否存在隐匿性的微侵袭癌灶的价值。结果表明,联合模型可明显提高DCIS患者升级风险的预测准确性,有助于让首次手术中应进行SLNB的高风险女性患者受益,而避免二次手术治疗,为临床治疗决策提供依据。

3.1 二元logistic回归分析DCISM的独立危险因素

       本研究中DCISM占比为28%,腋窝或前哨淋巴结转移的比率为2.5%,这与相关研究结果相似[26]。通过二元logistic回归分析,本研究发现病灶越大、核级别越高、伴有坏死、Ki-67高表达、P63阴性、存在可疑钙化以及最小ADC都是DCISM的独立危险因素,这与张敏等[2]的研究结果相似,该研究认为病变大小、核级别和淋巴结状态是DCISM的独立危险因素;PARK等[27]应用二元logistic回归分析也表明粗针穿刺、高核分级、腋窝淋巴结转移以及较大病变是DCIS升级的独立危险因素。但本研究中腋窝淋巴结状态并未作为独立危险因素出现,这一差异可能源于病理取材方法和患者筛选标准的不同。与既往研究不同,本研究筛选出的ADCmin是与DCISM相关的独立危险因素之一,采用SHAP分析各独立危险因素对准确诊断DCISM的贡献度,ADCmin也是既病灶最长径、Ki-67指数之后,位列第三位的主要因素,是临床医生术前评估DCISM风险的重要指标。最小ADC使用常规MRI设备自带软件即可获得,也为医生提供了一个简便、有效的预测工具。

3.2 MG特征与DCISM相关性

       MG摄影方面,本研究分析了DCIS和DCISM患者的腺体MG密度、MG特征、钙化状态及可疑钙化分布。结果发现,MG上仅表现为可疑钙化或可疑钙化伴或不伴其他特征,如不对称致密影、结构紊乱或肿块的病例中,DCSIM组均明显高于DCIS组,表明可疑钙化是DCISM最常见的MG表现,这与KIM等[28]的报道一致。本研究发现MG检查的可疑钙化与DCISM独立相关。HOU等[4]和LI等[29]利用乳腺MG特征表现为可疑钙化的图像构建了影像组学的预测模型,预测了经术前CNB诊断为DCIS患者的术后升级情况。本研究预测模型的构建方法虽然不同于以上研究,但研究结果有相似之处,均指出微钙化在DCIS伴浸润性疾病的术前诊断中起到了重要作用,可为临床医生选择最佳治疗策略提供依据。

3.3 乳腺MRI特征与DCISM相关性

       本研究分析评估了乳腺背景实质强化、病灶强化类型、病灶内部强化方式、TIC以及不同的ADC值在DCIS和DCISM之间的差异。其中非肿块样、不均匀强化及ADCmin在两组之间差异具有统计学意义。YOON等[30]认为,相比DCIS,DCISM中非肿块样强化更为常见,这与本研究结果一致。谭非易等[31]认为非肿块样强化,尤其是内部强化方式为廓清型曲线的簇环样强化对恶性病灶有较高的诊断价值,而本研究中DCISM组病灶内部不均匀强化率高于DCIS组。ADCmin与DCIS的升级有关。先前研究[32]表明ADCmin代表水分子扩散受限最明显的区域,也是肿瘤细胞增殖最活跃的区域、肿瘤细胞最密集的区域,与肿瘤的恶性程度相关,能够反映肿瘤的固有特性。本研究显示,DCIS与DCISM组之间的ADCmin存在显著差异,当该值小于/等于0.86×10-3 mm2/s时,DCIS发生微浸润的概率增高。这与SHAABAN等[33]的研究结果一致。

3.4 乳腺临床-病理及免疫组化与DCISM相关性

       DCIS与DCISM患者的年龄及月经状态无明显差异,但DCISM通常较DCIS表现出更具侵袭性的病理特征[33, 34],DCISM的病灶体积更大、核级别、坏死率、ER阴性率、HER-2阳性率、Ki-67指数以及P63阴性率较DCIS更高。研究[35]报道肿瘤体积越大,病变侵袭能力越强,这与本研究结果一致。本研究两组之间核级别差异具有统计学意义,高级别的DCIS在DCISM组中占主导地位,这与JIA等[36]的结果相似。肿瘤内的坏死是浸润性癌的常见特征,与临床预后和肿瘤转移有关[37, 38, 39]。在本研究中,DCISM组中坏死率占72%,显著高于纯DCIS组的33%(P<0.001),这表明肿瘤细胞的增殖和凋亡促进了坏死病灶的形成。此外,作为乳腺癌分子标志物的ER、PR、HER-2及Ki-67指数,其表达与癌症的形成、进展及预后密切相关。本研究发现DCISM组的ER阴性表达、HER-2阳性表达较DCIS组普遍,而PR表达在两组间未见明显差异,这与既往研究结果[2, 40]相似,表明ER阴性表达和HER-2阳性表达的DCIS更容易向侵袭性病变进展,而PR表达对提示DCIS是否具有侵袭性意义不大,但本研究多因素logistic回归分析显示ER和HER-2均不是DCISM的独立危险因素。Ki-67指数是反映肿瘤细胞增殖性的可靠标志物[41, 42],具有重要意义,本研究中DCISM的Ki-67指数明显高于DCIS,且差异具有统计学意义。肌上皮细胞通常介于管腔细胞和基底膜之间,作为防止管腔癌细胞侵入周围基底膜的物理屏障[43]。一项长期随访研究[44]发现,当肿瘤细胞随着肌上皮细胞层和基底膜的破坏而浸润周围基质后,未经治疗的DCIS病例最终进展为浸润性癌。此外,肌上皮细胞可能通过表达内源性蛋白酶抑制剂[45],诱导停滞肿瘤细胞周期,进而阻止癌细胞继续侵袭。因此,当肌上皮细胞缺失时,可高度提示DCIS升级为侵袭性病灶。而P63是识别肌上皮细胞的重要免疫标志物,在调节细胞增殖、分化和凋亡中起着关键作用[46]。P63的表达与乳腺癌的预后相关,高表达通常预后良好,反之则预后不良。在本研究中,DCISM组显示出更高的Ki-67表达和P63阴性率,Ki-67表达和P63状态是DCIS升级的独立危险因素。

3.5 本研究的局限性

       首先,本研究的样本量相对较小,并且数据来自单一机构。其次,由于研究人群均为有症状的患者,未对无症状的DCISM的相关性因素进行研究,均不可避免样本数据偏倚问题,今后需进一步大规模、多中心研究且纳入无症状DCISM患者避免此类问题发生。最后,本研究未应用人工智能和放射组学等较为先进、敏感的技术,未来需进一步探讨此类技术对诊断效能的影响。

4 结论

       总之,病灶最长径、核级别、坏死、Ki-67指数、P63状态、可疑钙化和ADCmin是DCIS伴微浸润的独立危险因素。本研究建立的联合模型对DCISM具有较高的预测效能,有望成为早期乳腺癌临床管理决策的影像生物标志工具并推广应用。

[1]
YOON G Y, CHOI W J, KIM H H, et al. Surgical outcomes for ductal carcinoma in situ: impact of preoperative MRI[J]. Radiology, 2020, 295(2): 296-303. DOI: 10.1148/radiol.2020191535.
[2]
张敏, 林青, 苏晓慧, 等. 乳腺导管原位癌伴微浸润与导管原位癌的临床病理及X线表现对比分析[J]. 中华放射学杂志, 2022, 56(2): 182-187. DOI: 10.3760/cma.j.cn112149-20210324-00268.
ZHANG M, LIN Q, SU X H, et al. Comparative analysis of clinicopathological and mammographic findings between ductal carcinoma in situ with microinvasion and ductal carcinoma in situ[J]. Chin J Radiol, 2022, 56(2): 182-187. DOI: 10.3760/cma.j.cn112149-20210324-00268.
[3]
GIULIANO A E, CONNOLLY J L, EDGE S B, 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. DOI: 10.3322/caac.21393.
[4]
HOU R, GRIMM L J, MAZUROWSKI M A, et al. Prediction of upstaging in ductal carcinoma in situ based on mammographic radiomic features[J]. Radiology, 2022, 303(1): 54-62. DOI: 10.1148/radiol.210407.
[5]
LEE S A, LEE Y, RYU H S, et al. Diffusion-weighted breast MRI in prediction of upstaging in women with biopsy-proven ductal carcinoma in situ[J/OL]. Radiology, 2022, 305(1): E60 [2023-08-14]. https://pubmed.ncbi.nlm.nih.gov/36154288/. DOI: 10.1148/radiol.229018.
[6]
CHOI S H, CHOI J S, HAN B K, et al. Long-term surveillance of ductal carcinoma in situ detected with screening mammography versus US: factors associated with second breast cancer[J]. Radiology, 2019, 292(1): 37-48. DOI: 10.1148/radiol.2019181844.
[7]
ROCHLIN D H, BARRIO A V, MCLAUGHLIN S, et al. Feasibility and clinical utility of prediction models for breast cancer-related lymphedema incorporating racial differences in disease incidence[J]. JAMA Surg, 2023, 158(9): 954-964. DOI: 10.1001/jamasurg.2023.2414.
[8]
CHE BAKRI N A, KWASNICKI R M, KHAN N, et al. Impact of axillary lymph node dissection and sentinel lymph node biopsy on upper limb morbidity in breast cancer patients: a systematic review and meta-analysis[J]. Ann Surg, 2023, 277(4): 572-580. DOI: 10.1097/SLA.0000000000005671.
[9]
李玉欣, 王嬴煊, 程流泉, 等. MRI、乳腺X线摄影和超声对乳腺导管原位癌检出的效能[J]. 中华放射学杂志, 2020, 54(6): 557-562. DOI: 10.3760/cma.j.cn112149-20200416-00564.
LI Y X, WANG Y X, CHENG L Q, et al. A retrospective comparison of MRI, ultrasound and X-ray mammography in detecting breast ductal carcinoma in situ[J]. Chin J Radiol, 2020, 54(6): 557-562. DOI: 10.3760/cma.j.cn112149-20200416-00564.
[10]
KEYMEULEN K B I M, GEURTS S M E, KOOREMAN L F S, et al. Clinical value of contralateral breast cancers detected by pre-operative MRI in patients diagnosed with DCIS: a population-based cohort study[J]. Eur Radiol, 2023, 33(3): 2209-2217. DOI: 10.1007/s00330-022-09115-5.
[11]
LEE K H, HAN J W, KIM E Y, et al. Predictive factors for the presence of invasive components in patients diagnosed with ductal carcinoma in situ based on preoperative biopsy[J/OL]. BMC Cancer, 2019, 19(1): 1201 [2023-08-14]. https://pubmed.ncbi.nlm.nih.gov/31822268/. DOI: 10.1186/s12885-019-6417-3.
[12]
YOON G Y, CHOI W J, CHA J H, et al. The role of MRI and clinicopathologic features in predicting the invasive component of biopsy-confirmed ductal carcinoma in situ[J/OL]. BMC Med Imaging, 2020, 20(1): 95 [2023-08-14]. https://pubmed.ncbi.nlm.nih.gov/32787871/. DOI: 10.1186/s12880-020-00494-z.
[13]
CHOU S S, ROMANOFF J, LEHMAN C D, et al. Preoperative breast MRI for newly diagnosed ductal carcinoma in situ: imaging features and performance in a multicenter setting (ECOG-ACRIN E4112 trial)[J/OL]. Radiology, 2021, 301(1): E381 [2023-08-14]. https://pubmed.ncbi.nlm.nih.gov/34543146/. DOI: 10.1148/radiol.2021219016.
[14]
CHEN H, LI X, LI F L, et al. Prediction of coexisting invasive carcinoma on ductal carcinoma in situ (DCIS) lesions by mass spectrometry imaging[J]. J Pathol, 2023, 261(2): 125-138. DOI: 10.1002/path.6154.
[15]
王雯琼, 杨蔚, 刘开惠, 等. 乳腺X线摄影、MRI及病理联合诊断乳腺癌分子分型[J]. 中国医学影像技术, 2023, 39(2): 210-215. DOI: 10.13929/j.issn.1003-3289.2023.02.013.
WANG W Q, YANG W, LIU K H, et al. Combination of mammography, MRI and pathology for diagnosis of molecular subtypes of breast cancer[J]. Chin J Med Imag Technol, 2023, 39(2): 210-215. DOI: 10.13929/j.issn.1003-3289.2023.02.013.
[16]
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.
[17]
金哲, 张斌, 张璐, 等. 基于多参数MRI影像组学及临床特征的鼻咽癌远处转移可解释性机器学习预测模型[J]. 磁共振成像, 2022, 13(11): 22-29. DOI: 10.12015/issn.1674-8034.2022.11.005.
JIN Z, ZHANG B, ZHANG L, et al. Prediction of distant metastasis in nasopharyngeal carcinoma by interpretable machine learning model based on multiparametric MRI radiomics and clinical factors[J]. Chin J Magn Reson Imag, 2022, 13(11): 22-29. DOI: 10.12015/issn.1674-8034.2022.11.005.
[18]
SPAK D A, PLAXCO J S, SANTIAGO L, et al. BI-RADS® fifth edition: a summary of changes[J]. Diagn Interv Imaging, 2017, 98(3): 179-190. DOI: 10.1016/j.diii.2017.01.001.
[19]
VAN DER HOOGT K J J, SCHIPPER R J, WESSELS R, et al. Breast DWI analyzed before and after gadolinium contrast administration-an intrapatient analysis on 1.5 T and 3.0 T[J]. Invest Radiol, 2023, 58(12): 832-841. DOI: 10.1097/RLI.0000000000000999.
[20]
KIM J J, KIM J Y, SUH H B, 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): 822-833. DOI: 10.1007/s00330-021-08166-4.
[21]
LI Y B, CHEN Y Z, ZHAO R, et al. Development and validation of a nomogram based on pretreatment dynamic contrast-enhanced MRI for the prediction of pathologic response after neoadjuvant chemotherapy for triple-negative breast cancer[J]. Eur Radiol, 2022, 32(3): 1676-1687. DOI: 10.1007/s00330-021-08291-0.
[22]
WANG L J, CHANG L F, LUO R, et al. An artificial intelligence system using maximum intensity projection MR images facilitates classification of non-mass enhancement breast lesions[J]. Eur Radiol, 2022, 32(7): 4857-4867. DOI: 10.1007/s00330-022-08553-5.
[23]
LUNKIEWICZ M, FORTE S, FREIWALD B, et al. Interobserver variability and likelihood of malignancy for fifth edition BI-RADS MRI descriptors in non-mass breast lesions[J]. Eur Radiol, 2020, 30(1): 77-86. DOI: 10.1007/s00330-019-06312-7.
[24]
王英哲, 殷咏梅, 江泽飞. 2023年CSCO《乳腺癌诊疗指南》更新要点解读[J]. 中国肿瘤外科杂志, 2023, 15(3): 209-213, 218. DOI: 10.3969/j.issn.1674-4136.2023.03.001.
WANG Y Z, YIN Y M, JIANG Z F. Interpretation of updated key points of Chinese society of clinical oncology (CSCO) breast cancer guidelines 2023[J]. Chin J Surg Oncol, 2023, 15(3): 209-213, 218. DOI: 10.3969/j.issn.1674-4136.2023.03.001.
[25]
LANDIS J R, KOCH G G. The measurement of observer agreement for categorical data[J]. Biometrics, 1977, 33(1): 159-174.
[26]
李霖, 欧阳汝珊, 林小慧, 等. 基于深度学习乳腺X线摄影联合自然语言处理预测不同病理进展期乳腺导管原位癌预后研究[J]. 中华放射学杂志, 2022, 56(11): 1215-1222. DOI: 10.3760/cma.j.cn112149-20220321-00258.
LI L, OUYANG R S, LIN X H, et al. A study on the prediction of prognosis of ductal carcinoma in situ at different pathological stages based on deep learning mammography combined with natural language processing[J]. Chin J Radiol, 2022, 56(11): 1215-1222. DOI: 10.3760/cma.j.cn112149-20220321-00258.
[27]
PARK K W, KIM S W, HAN H, et al. Ductal carcinoma in situ: a risk prediction model for the underestimation of invasive breast cancer[J/OL]. NPJ Breast Cancer, 2022, 8(1): 8 [2023-08-14]. https://pubmed.ncbi.nlm.nih.gov/35031626/. DOI: 10.1038/s41523-021-00364-z.
[28]
KIM M, KIM H J, CHUNG Y R, et al. Microinvasive carcinoma versus ductal carcinoma In situ: a comparison of clinicopathological features and clinical outcomes[J]. J Breast Cancer, 2018, 21(2): 197-205. DOI: 10.4048/jbc.2018.21.2.197.
[29]
LI J, SONG Y, XU S Y, et al. Predicting underestimation of ductal carcinoma in situ: a comparison between radiomics and conventional approaches[J]. Int J Comput Assist Radiol Surg, 2019, 14(4): 709-721. DOI: 10.1007/s11548-018-1900-x.
[30]
YOON G Y, CHOI W J, KIM H H, et al. Outcomes and imaging features of microinvasive carcinoma and ductal carcinoma in situ: matched cohort study[J/OL]. Clin Imaging, 2023, 96: 64-70 [2023-08-14]. https://pubmed.ncbi.nlm.nih.gov/36827842/. DOI: 10.1016/j.clinimag.2023.01.004.
[31]
谭非易, 刘梅, 程流泉, 等. 乳腺MRI簇环样强化的影像与病理对照研究[J]. 中国医学影像学杂志, 2020, 28(8): 596-601. DOI: 10.3969/j.issn.1005-5185.2020.08.010.
TAN F Y, LIU M, CHENG L Q, et al. Imaging-pathological correlation of clustered ring sign on breast MRI[J]. Chin J Med Imag, 2020, 28(8): 596-601. DOI: 10.3969/j.issn.1005-5185.2020.08.010.
[32]
ZHAO S H, SHAO G R, CHEN P P, et al. Diagnostic performance of minimum apparent diffusion coefficient value in differentiating the invasive breast cancer and ductal carcinoma in situ[J]. J Cancer Res Ther, 2019, 15(4): 871-875. DOI: 10.4103/jcrt.JCRT_607_18.
[33]
SHAABAN A M, HILTON B, CLEMENTS K, et al. The presentation, management and outcome of patients with ductal carcinoma in situ (DCIS) with microinvasion (invasion ≤1 mm in size)-results from the UK Sloane Project[J]. Br J Cancer, 2022, 127(12): 2125-2132. DOI: 10.1038/s41416-022-01983-4.
[34]
LEE S Y, YOO T K, KIM J, et al. Clinical significance of microinvasive breast cancer across the different subtypes and human epidermal growth factor receptor 2 expression levels[J]. Breast Cancer Res Treat, 2023, 200(1): 47-61. DOI: 10.1007/s10549-023-06955-7.
[35]
ZHAO M R, MA W J, SONG X C, et al. Feasibility analysis of magnetic resonance imaging-based radiomics features for preoperative prediction of nuclear grading of ductal carcinoma in situ[J]. Gland Surg, 2023, 12(9): 1209-1223. DOI: 10.21037/gs-23-132.
[36]
JIA H Q, ZHAO P, CHEN Z X, et al. Clinicopathological characteristics and prognostic analysis of tumor-infiltrating lymphocytes (TILs) in ductal carcinoma in situ (DCIS) and DCIS with microinvasion (DCIS-Mi) of the breast[J]. Breast Cancer Res Treat, 2022, 193(1): 111-120. DOI: 10.1007/s10549-022-06553-z.
[37]
IZQUIERDO-ALTAREJOS P, MORENO-MANZANO V, FELIPO V. Pathological and therapeutic effects of extracellular vesicles in neurological and neurodegenerative diseases[J]. Neural Regen Res, 2024, 19(1): 55-61. DOI: 10.4103/1673-5374.375301.
[38]
TANG H L, CAO C H, ZHANG G Y, et al. Impact of particle size of multivesicular liposomes on the embolic andtherapeutic effects in rabbit VX2 liver tumor[J]. Drug Deliv, 2023, 30(1): 1-16. DOI: 10.1080/10717544.2022.2157519.
[39]
LU Z C, TANG F C, LI Z C, et al. Prognosis risk model based on pyroptosis-related lncRNAs for bladder cancer[J/OL]. Dis Markers, 2022, 2022: 7931393 [2023-08-14]. https://pubmed.ncbi.nlm.nih.gov/35154513/. DOI: 10.1155/2022/7931393.
[40]
ZHENG J, ZHOU T, LI F, et al. Clinic-pathological features of breast ductal carcinoma in situ with micro-invasion[J]. Cancer Invest, 2020, 38(2): 113-121. DOI: 10.1080/07357907.2020.1715422.
[41]
刘宏, 刘显旺, 刘光耀, 等. 表观扩散系数术前区分Luminal型与非Luminal型乳腺癌及其与Ki-67增殖指数的相关性研究[J]. 磁共振成像, 2023, 14(4): 51-56. DOI: 10.12015/issn.1674-8034.2023.04.010.
LIU H, LIU X W, LIU G Y, et al. The value of apparent diffusion coefficient value in differentiating the Luminal-type and non-Luminal-type breast cancer and evaluating tumor cell proliferation activity[J]. Chin J Magn Reson Imag, 2023, 14(4): 51-56. DOI: 10.12015/issn.1674-8034.2023.04.010.
[42]
SMITS H J G, RUITER L N, BREIMER G E, et al. Using intratumor heterogeneity of immunohistochemistry biomarkers to classify laryngeal and hypopharyngeal tumors based on histologic features[J/OL]. Mod Pathol, 2023, 36(8): 100199 [2023-08-14]. https://pubmed.ncbi.nlm.nih.gov/37116830/. DOI: 10.1016/j.modpat.2023.100199.
[43]
RISOM T, GLASS D R, AVERBUKH I, et al. Transition to invasive breast cancer is associated with progressive changes in the structure and composition of tumor stroma[J/OL]. Cell, 2022, 185(2): 299-310.e18 [2023-08-14]. https://pubmed.ncbi.nlm.nih.gov/35063072/. DOI: 10.1016/j.cell.2021.12.023.
[44]
DECKWIRTH V, RAJAKYLÄ E K, CATTAVARAYANE S, et al. Cytokeratin 5 determines maturation of the mammary myoepithelium[J/OL]. iScience, 2021, 24(5): 102413 [2023-08-14]. https://pubmed.ncbi.nlm.nih.gov/34007958/. DOI: 10.1016/j.isci.2021.102413.
[45]
GIBSON S V, TOMAS BORT E, RODRÍGUEZ-FERNÁNDEZ L, et al. TGFβ-mediated MMP13 secretion drives myoepithelial cell dependent breast cancer progression[J/OL]. NPJ Breast Cancer, 2023, 9(1): 9 [2023-08-14]. https://pubmed.ncbi.nlm.nih.gov/36864079/. DOI: 10.1038/s41523-023-00513-6.
[46]
HUANG C Z, ZHAO S M, YANG Y J, et al. TP63 gain-of-function mutations cause premature ovarian insufficiency by inducing oocyte apoptosis[J/OL]. J Clin Invest, 2023, 133(5): e162315 [2023-08-14]. https://pubmed.ncbi.nlm.nih.gov/36856110/. DOI: 10.1172/JCI162315.

上一篇 基于DCE-MRI的3D-MIP重建及多参数评估BI-RADS 4类乳腺肿瘤
下一篇 基于术前MRI特征及定量参数列线图预测乳腺癌患者淋巴血管侵犯
  
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