乳腺病变术前MRI误诊分析

Analysis of cases with MRI misdiagnosed breast lesions

目的:分析乳腺病变术前MRI漏诊、高估、低估病例的临床、影像及病理特点,探讨MRI误诊原因。方法:回顾性分析行乳腺MRI检查并经手术病理证实的175名患者,共计222枚病灶的临床、影像及病理资料,分析其MRI表现特点并按BI-RADS分类进行评估;比较MRI对不同强化类型病灶诊断效能。以患者临床信息及BI-RADS-MRI词典为依据采用二元logistic回归分析建立模型预测病灶良恶性,并与评估结果对比,比较不同影像学表现对诊断结果的影响。结果:222枚病灶中良性161个、恶性61个。MRI漏诊6个病理均为良性;低估4个病灶中3个含导管原位癌成分;高估14个病灶包括12个肿块型和2个非肿块强化病灶。ADC值对肿块型病灶诊断效能高、AUC达0.932,对非肿块强化病灶诊断效能较差;DCE-MRI联合DWI诊断乳腺癌的敏感度93.4%,特异度91.3%,诊断准确度91.9%;MRI对肿块型和非肿块强化病灶总体诊断效能无显著差异。结论:乳腺DCE-MRI联合DWI诊断乳腺恶性肿瘤的敏感度、特异度均较高,对肿块样强化与非肿块强化病灶的总体诊断效能相仿。低估的病灶病理以导管原位癌为主。医师进行BI-RADS-MRI分类评估时带有一定主观性,诊断结果有高估的趋势、同时减少大部分低估,对总体诊断准确性影响不大。

Objective：Joint analysis of clinical,MRI,and pathological features of missed,false-negative,and false-posi-tive lesions was performed to explore the causes of preoperative misdiagnosis.Methods：A consecutive cohort of 175 patients who underwent preoperative breast MRI with 222 pathologically confirmed lesions were retrospectively reviewed.Two breast imaging radiologists performed the MRI interpretation according to BI-RADS atlas.Comparison of MRI diagnostic performance between different lesion types was carried out.Logistic regression models were also established to differentiate benign and malignant lesions.Results：Of all 222 breast lesions （161 benign and 61 malignant）,6 benign lesions were missed on MRI.Three in 4 false-negative lesions had ductal carcinoma in situ （DCIS）component.The other 14 lesions were overes-timated.ADC value worked great for masses with AUC of 0.932,but poor for non-mass enhanced NME lesions.The sensi-tivity,specificity,and accuracy of DCE-MRI combined with DWI were 93.4%,91.3%,and 91.9%,respectively.The overall accuracy of mass and NME lesions was similar.Conclusion：Accuracy of DCE-MRI combined with DWI was high in differen-tiating benign and malignant lesions,and similar in masses and NME lesions.Underestimated lesions were mainly DCIS.Ra-diologists tended to overestimate lesions and avoided underestimation at the same time,in that the overall accuracy was al-most unaffected.