基于深度学习技术对三阴性乳腺癌的多模态影像学研究

Multimodal Imaging Study of Triple Negative Breast Carcinoma Based on Deep Learning Technology

目的 对比研究数字乳腺X线摄影(DM)在联合人工智能诊断系统(AI)后与超声(US)、动态增强MRI(DCE-MRI)等影像学检查对三阴性乳腺癌(TNBC)的诊断效能,并评估AI的作用。方法 收集TNBC 42例,乳腺良性病例68例,所有病例均有完整术前的DM、US和DCE-MRI检查资料。回顾乳腺DM、US与DCE-MRI影像资料,各得出一组诊断结果;另由AI、AI联合医师分别对DM图像进行诊断,得出两组结果。对比分析5种诊断方法的受试者工作特征曲线、曲线下面积(AUC)、特异度、敏感度、阳性预测值和阴性预测值,并对比各诊断方法的一致性。结果 各诊断方法的AUC值中DM(医师)组最低,而DM(AI+医师)组AUC值最高,为0.770,优于US、DCE-MRI;DCE-MRI与US检查的敏感度最高,但其特异度较差。结论 此研究中DM(医师)联合AI可以有效提高其对三阴性乳腺癌的诊断效能,其综合诊断效能略优于磁共振检查。

Objective To compare the diagnostic efficacy of digital mammography(DM) combined with artificial intelligence diagnostic system(AI) with ultrasound(US) and dynamic enhanced MRI(DCE-MRI) in the diagnosis of triple-negative breast neoplasms(TNBC),and evaluate the role of AI. Methods 42 cases of TNBC and 68 cases of benign breast cancer were collected. All cases had complete preoperative data of DM,US and DCE-MRI. The data of DM,US and DCE-MRI of breast were reviewed,and a set of diagnostic results were obtained respectively. In addition,the DM images were diagnosed by AI and AI joint radiologist respectively,and two sets of results were obtained. The subject operating characteristic curves,area under curve(AUC),specificity,sensitivity,positive predictive value and negative predictive value of the five diagnostic methods were compared and analyzed,and the consistency of each diagnostic method was compared. Results The AUC value of DM(radiologist) group was the lowest,while the AUC value of DM(AI+ radiologist) group was the highest,which was 0.770,better than US and DCE-MRI. The sensitivity of DCE-MRI and US examination was the highest,but its specificity was poor. Conclusion In the study,DM(radiologist) combined with AI can effectively improve its diagnostic efficacy in triple-negative breast cancer,and its comprehensive diagnostic efficacy is slightly better than that of MRI.