| 摘要: |
| 目的 探讨超声人工智能（AI）辅助诊断系统对最大径≤2 cm的乳腺影像报告与数据系统（BI-RADS）4类乳腺结节的诊断价值。方法 回顾性分析2020年5月至2022年10月于上海中医药大学附属曙光医院进行超声检查并诊断为BI-RADS 4类乳腺结节的204例患者共210个最大径≤2 cm结节的二维超声图像。以术后病理结果为金标准，评价常规超声和AI系统（风险评分值阈值设为0.65、0.70）对最大径≤2 cm的BI-RADS 4类乳腺结节良恶性的诊断效能。结果 210个乳腺结节中良性结节94个，恶性结节116个。高年资超声医师常规超声检查诊断乳腺结节良恶性的灵敏度为92.24%，特异度为75.53%，准确度为84.76%；AI系统（阈值0.65）诊断乳腺结节良恶性的灵敏度为92.24%，特异度为71.28%，准确度为82.86%；AI系统（阈值0.70）诊断乳腺结节良恶性的灵敏度为90.52%，特异度为79.79%，准确度为85.71%。AI系统（阈值0.70）诊断BI-RADS 4a类结节的准确度高于常规超声和AI系统（阈值0.65）（79.41% vs 77.94%、75.00%）。高年资超声医师通过常规超声对最大径≤1 cm的结节诊断准确度最高，为86.36%，AI系统（阈值0.65）及AI系统（阈值0.70）准确度分别为81.82%、84.09%。结论 超声AI辅助诊断系统可辅助鉴别诊断最大径≤2 cm的BI-RADS 4类乳腺结节的良恶性。 |
| 关键词: 超声检查 人工智能 计算机辅助诊断系统 乳腺肿瘤 乳腺影像报告与数据系统 |
| DOI：10.16781/j.CN31-2187/R.20230009 |
| 投稿时间：2023-01-12修订日期：2023-11-09 |
| 基金项目:上海市虹口区卫生健康委员会医学科研课题（虹卫2302-26），上海市虹口区卫生健康委员会临床重点扶持专科项目（HKLCFC202404），海军军医大学（第二军医大学）第二附属医院人才建设三年行动计划——“金字塔人才工程”军事医学人才项目（1009），同济大学附属上海市第四人民医院科研启动专项（SYKYQD06101）．Supported by Medical Research Project of Health Commission of Shanghai Hongkou District (HW2302-26), Clinical Key Supporting Project of Health Commission of Shanghai Hongkou District (HKLCFC202404), Military Medical Talent Project of “Pyramid Talent Program” of Three-Year Action Plan for Talent Construction of The Second Affiliated Hospital of Naval Medical University (Second Military Medical University) (1009), Start-up Scientific Research Project of Shanghai Fourth People’s Hospital Affiliated to Tongji University (SYKYQD06101). |
|
| Ultrasonic artificial intelligence system in diagnosis of BI-RADS class 4 breast nodules with maximum diameter ≤2 cm |
| CHEN Rui1,WU Shu1,GUO Jia1,GUO Fangqi2,ZHAO Jiaqi2* |
(1. Department of Ultrasound, Shuguang Hospital Affiliated to Shanghai University of Traditional Chinese Medicine, Shanghai 201203, China;
2. Department of Ultrasound, Shanghai Fourth People's Hospital Affiliated to Tongji University, Shanghai 200434, China
* Corresponding author) |
| Abstract: |
| Objective To evaluate the diagnostic value of ultrasonic artificial intelligence (AI) system in breast imaging-reporting and data system (BI-RADS) class 4 breast nodules with maximum diameter ≤2 cm. Methods A total of 210 breast nodules were analyzed in 204 patients with pathological results obtained by surgery at Shuguang Hospital Affiliated to Shanghai University of Traditional Chinese Medicine. The maximum diameter of the breast nodules was ≤2 cm. With the pathological results as the gold standard, all nodules were examined by routine ultrasound (US) and AI aided diagnosis system. The values of routine US by a senior physocian, AI (threshold value 0.65) and AI (threshold value 0.70) in diagnosing benign and malignant in BI-RADS class 4 breast nodules with maximum diameter ≤2 cm were evaluated. Results The pathological results showed that 210 breast nodules included 94 benign nodules and 116 malignant nodules. The sensitivity, specificity and accuracy of routine US in diagnosing benign and malignant breast nodules were 92.24%, 75.53% and 84.76%, respectively; the sensitivity, specificity and accuracy of AI (threshold value 0.65) were 92.24%, 71.28% and 82.86%, respectively; and those of AI (threshold value 0.70) were 90.52%, 79.79% and 85.71%, respectively. The accuracy of diagnosing BI-RADS 4a nodules of AI (threshold value 0.70) was higher than those of routine US and AI (threshold value 0.65) (79.41% vs 77.94%, 75.00%). The senior physician had the highest diagnostic accuracy of 86.36% for nodules with maximum diameter ≤1 cm using routine US. The accuracies of the AI system with threshold value 0.65 and 0.70 were 81.82% and 84.09%, respectively. Conclusion Ultrasonic AI diagnosis system can assist to improve the diagnostic efficiency of BI-RADS class 4 breast nodules with maximum diameter ≤2 cm. |
| Key words: ultrasonography artificial intelligence computer aided diagnosis system breast neoplasms breast imaging-reporting and data system |
