人工智能辅助低年资放射科医师检出乳腺病变

目的观察人工智能(AI)辅助对低年资放射科医师检出乳腺X线片中乳腺病灶的价值。方法回顾性分析73例接受乳腺X线摄影检查的女性患者。由3名低年资放射科医师首先分别阅片,之后于AI辅助下再次阅片,勾画乳腺病灶区域;统计检出病灶数,记录病灶类型及乳腺影像报告和数据系统(BI-RADS)分类。以病理结果或随访最终临床诊断为标准,对比医师单独阅片与AI辅助阅片检出乳腺病灶的敏感度和准确率及对不同类型、不同BI-RADS分类病灶的诊断效能。结果 73例共121个乳腺病灶,其中软组织病灶(包括肿块、结构扭曲与不对称病灶)88个、钙化病灶33个;BI-RADS 2类17个,3类66个,4类及以上38个。与医师单独阅片相比,AI辅助阅片对全部乳腺病灶、特别是软组织病灶的诊断敏感度和准确率均显著提升(P均0.01),而对乳腺钙化病灶的敏感度和准确率差异均无统计学意义(P均0.05)。结论 AI辅助有助于提高低年资放射科医师检出乳腺X线片中乳腺病灶、尤其软组织病灶的效能。

Artificial intelligence assisted junior radiologists in detecting breast lesions on mammography

Objective To observe the application value of artificial intelligence (AI) for junior radiologists in detecting breast lesions on mammography. Methods Seventy-three women who underwent mammography were retrospectively analyzed. The mammographic images reviewed were interpreted by junior radiologists, respectively. Then the second reviews were performed with the assistance of AI, and the breast lesion areas were delineated, respectively, the numbers of detected lesions were counted, and the types of lesions and the classifications of breast imaging report and data system (BI-RADS) were evaluated. Taken pathological results or the final clinical diagnosis after following-up as the standards, the sensitivity and accuracy of detecting breast lesions, as well as the diagnostic efficacies of lesions with different types and different BI-RADS grades were compared between junior radiologists alone and with the assistance of AI. Results A total of 121 lesions were identified in 73 patients, including 88 of tissue lesions (mass, structural distortion or asymmetry) and 33 calcifications. According to BI-RADS classification system, there were 17 grade 2 lesions, 66 grade 3 lesions and 38 grade 4 and above lesions. Compared with junior radiologists alone, AI-assistant interpretation significantly improved the overall diagnostic sensitivity and accuracy of all 121 breast lesions, especially of soft tissue lesions (all P<0.01), while there was no statistical significance of sensitivity nor accuracy of diagnosing breast calcifications with or without AI assistance (all P>0.05). Conclusion AI assistance could improve the efficiency of junior radiologists for detecting breast lesions on mammography, especially of soft tissue lesions.