乳腺CEUS预测模型诊断乳腺恶性病灶的观察者间一致性:多中心研究

摘要

目的 评价采用乳腺CEUS预测模型诊断乳腺恶性病灶的观察者一致性.方法 收集多中心共953例接受超声和CEUS检查的乳腺单发结节患者.本课题组由初始组(各医院1名低年资医师)、检查者组(各医院1～2名具有2年以上CEUS检查经验的医师)、研究组(四川省医学科学院四川省人民医院2名高年资医师)及交叉盲读组(各医院1～2名副主任医师或主任医师)构成.首先由初始组及检查者组根据乳腺影像报告和数据系统(BI-RADS)对病灶进行分类,其次由交叉盲读组和研究组采用乳腺CEUS预测模型再次进行BI-RADS分类.以病理结果为金标准,计算4组诊断乳腺恶性病灶的效能;分析观察者间诊断乳腺恶性病灶的一致性.结果 953例中,病理证实良性病灶451例(451/953,47.32％),恶性病灶435例(435/953,45.65％),癌前病变67例(67/953,7.03％).初始组、检查者组、研究组及交叉盲读组诊断乳腺恶性病灶的准确率分别为71.67％(683/953)、74.92％ (714/953)、80.17％(764/953)及83.42％ (795/953).初始组与检查者组诊断乳腺恶性病灶的一致性较好(Kappa=0.82,P＜0.001),与研究组及交叉盲读组的一致性中等(Kappa=0.56、0.41,P均＜0.001);检查者组与研究组、交叉盲读组的一致性均为中等(Kappa=0.68、0.51,P均＜0.001);研究组与交叉盲读组的一致性中等(Kappa=0.74,P＜0.001).结论 不同观察者采用乳腺CEUS预测模型诊断乳腺恶性病灶的一致性一般.%Objective To explore the consistency of different physicians in diagnosis of malignant breast lesions with breast CEUS predictive model.Methods Totally 953 patients with solitary breast nodule from multicenter who underwent ultrasound and CEUS were collected.The research team was composed by the initial group (one junior physician from each hospital),check group (one or two physicians who had at least two-year experience of CEUS examination from each hospital),research group (two senior physicians from Sichuan Academy of Medical Sciences & Sichuan Provincial People's Hospital) and cross-blinded group (one or two vice directors or chief physicians from each hospital).At first,the lesions were classified according to the breast imaging reporting and data system (BI RADS) by the initial group and the check group,then new BI RADS classifications were performed by research group and cross blind group with breast CEUS predictive model.The consistency of different physicians in diagnosis of malignant breast lesions was analyzed.Results Among 953 patients,benign lesions were found in 451 patients (451/953,47.32％),malignant lesions were found in 435 patients (435/953,45.65％),and precancerous lesions were found in 67 patients (67/953,7.03％).The accuracy of the initial group,check group,research group and cross-blinded group was 71.67％(683/953),74.92％(714/953),80.17％ (764/953) and 83.42 ％ (795/953),respectively.The consistency of different physicians for diagnosis of malignant breast lesions between initial group and check group was good (Kappa =0.82,P＜0.001),while between initial group and crossblinded group,initial group and research group were both moderate (Kappa =0.56,0.41;all P＜0.001).The consistency of different physicians for diagnosis of malignant breast lesions between cheek group and cross-blinded group,between check group and research group were both moderate (Kappa =0.68,0.51;all P＜0.001).The consistency between research group and cross-blinded group with breast CEUS predictive model was moderate (Kappa =0.74,P＜ 0.001).Conclusion The consistency of different physicians in diagnosis of malignant breast lesions with breast CEUS predictive model was moderate.