基于Shapley additive explanations(SHAP)分析乳腺X线检查假阴性乳腺癌的特征  

Study of Features of False⁃negative Mammography Breast Cancer Based on the Analysis of Shapley Additive Explanations(SHAP)

作　　者：杨蔚[1] 张宁妹[2] 张朝林[3] 尹清云[4] YANG Wei;ZHANG Ningmei;ZHANG Chaolin;YIN Qingyun(Department of Radiology,General Hospital of Ningxia Medical University;Department of Pathology,General Hospital of Ningxia Medical University;Department of Surgical Oncology,General Hospital of Ningxia Medical University;Department of Medical Oncology,General Hospital of Ningxia Medical University)

机构地区：[1]宁夏医科大学总医院放射科 [2]宁夏医科大学总医院病理科 [3]宁夏医科大学总医院肿瘤外三科 [4]宁夏医科大学总医院肿瘤内三科

出　　处：《中国医学计算机成像杂志》2023年第3期271-276,共6页Chinese Computed Medical Imaging

基　　金：2022年宁夏回族自治区重点研发计划项目(2022BEG03166)。

摘　　要：目的:基于Shapley additive explanations(SHAP)分析乳腺X线摄影(MG)检查假阴性乳腺癌的特征。方法:回顾性收集我院2018年1月—2022年5月首诊乳腺癌患者,评估临床病理和影像学表现。通过logistic回归分析确定MG假阴性乳腺癌的独立危险因素。应用SHAP分析各独立危险因素的贡献价值。结果:1505个病灶纳入研究,MG假阴性乳腺癌373个。较小的病灶、致密或不均匀致密的MG密度、无可疑钙化、非肿块样强化(NME)、原位癌和腋窝淋巴结阴性与MG假阴性乳腺癌独立相关,上述参数构建的联合模型具有良好的预测效能(AUC=0.871)。根据SHAP分析,病灶大小、MG密度和病灶MRI强化类型是导致MG假阴性的主要因素;钙化状态、病理类型和腋窝淋巴结状态是重要补充。结论:MG假阴性乳腺癌具有一定的特征,此类患者需补充乳腺MRI检查以提高癌灶的检出率。Purpose:To identify factors independently associated with only MRI-detected breast cancer leading to false-negative mammography based on Shapley additive explanations(SHAP).Methods:Between January 2018 and May 2022,the consecutive breast cancer patients with mammography(MG)and magnetic resonance imaging(MRI)were retrospectively reviewed for the false-negative breast cancer on MG.Clinicopathological features and imaging findings were assessed.Logistic regression analysis was performed to identify the independent risk factors associating with MG false-negative breast cancer.The SHAP was applied to determine the contribution of each independent risk factor.Results:Totally 1505 malignant lesions met the inclusion criteria.There were 373 MG falsenegative breast cancers.Small lesion,extreme or heterogeneous mammographic density,absence of suspicious calcification,non-mass enhancement(NME),carcinoma in situ and axillary lymph node negative were independent risk factors associated with MG false-negative breast cancer.The combined model including the above parameters had a good predictive performance(AUC=0.871).According to SHAP value,lesion size,mammographic density and enhancement type on MRI were the most significant factors,while calcification status,histological type,and axillary lymph node status were the supplemental contributors for MG false-negative breast cancer.Conclusions:There are some features in MG false-negative breast cancer,which should be considered to perform breast MRI to improve the detection rate of this type of cancer.

关 键 词：乳腺肿瘤 X线摄影 磁共振成像 假阴性 SHAP 

分 类 号：R445.4[医药卫生—影像医学与核医学]