中国实用外科杂志

中国实用外科杂志

• 论蓍 • 上一篇    下一篇

人工智能影像辅助诊断平台对直肠癌壁外血管侵犯识别多中心临床研究

刘书豪1a，苏柯帆1a，张宪祥1a，刘广伟1a，张正东2，马金龙1b，周晓明1b，李    硕1a，信芳杰1c，王继刚1c，姚宏伟3，王贵英4，肖    刚5，袁维堂6，康    亮7，张佃良8，李    帅2，卢    云1a，9   

  1. 1青岛大学附属医院 a.普通外科医疗中心  b.影像科 c.病理科，山东青岛266003；2北京航空航天大学虚拟现实技术与系统国家重点实验室，北京100191；3首都医科大学附属北京友谊医院普外科 国家消化系统疾病临床医学研究中心，北京 100050；4河北医科大学第四医院普通外科，河北石家庄050011；5北京医院普通外科  国家老年医学中心，北京100730；6郑州大学第一附属医院普通外科，郑州河南450052；7中山大学附属第六医院普通外科，广东广州510655；8青岛市立医院普通外科，山东青岛266011；9山东省数字医学与计算机辅助手术重点实验室，山东青岛266003
  • 出版日期:2019-10-01 发布日期:2019-10-10

  • Online:2019-10-01 Published:2019-10-10

摘要：

目的    探索基于区域卷积神经网络（Faster R-CNN）的人工智能影像辅助诊断平台对识别直肠癌壁外血管侵犯（EMVI）的临床应用价值。方法    回顾性分析我国7家医院2016年7月至2019年2月行直肠高分辨MRI检查的500例直肠癌病人的临床资料，其中EMVI阳性174例和EMVI阴性326例。使用随机数字法将病人资料随机分配到训练组（400例，包括EMVI阳性133例和EMVI阴性267例）和验证组（100例，包括EMVI阳性41例和EMVI阴性59例）。利用Faster R-CNN对训练组20 430张高分辨MRI图像进行学习和训练，建立人工智能影像辅助诊断平台。对验证组5107张高分辨MRI图像进行临床验证，应用受试者工作特征曲线（ROC）和曲线下面积（AUC），比较人工智能影像辅助诊断平台与高年资影像专家的诊断结果。结果    人工智能影像辅助诊断平台识别EMVI的准确率、敏感度、特异度、阳性预测值和阴性预测值分别为93.4%、97.3%、89.5%、0.90和0.97。AUC为0.98。自动识别单张图像所需时间为0.2 s，而影像学专家判读时间约为10 s。结论    基于Faster R-CNN的人工智能影像辅助诊断平台对识别直肠癌EMVI具有高效率和可行性，可辅助影像科医生进行成像诊断。

关键词: 人工智能, 直肠癌, 壁外血管侵犯, 磁共振成像

Abstract:

Faster R-CNN-based artificial intelligence image-aided diagnosis platform in identifying EMVI of rectal cancer:  A multicenter clinical study        LIU Shu-hao*，SU Ke-fan，ZHANG Xian-xiang，et al. *Department of General Surgery，Affiliated Hospital of Qingdao University，Qingdao 266003，China
Corresponding author：LU Yun，E-mail：cloudylucn@126.com
Abstract    Objective    To explore the clinical application value of artificial intelligence image aided diagnosis platform based on Faster R-CNN in identifying EMVI of rectal cancer. Methods    In the multicenter retrospective study，500 patients with rectal cancer who underwent high-resolution MRI examination between July 2016 and February 2019 were selected from seven hospitals in China. They were divided into 174 positive and 326 negative patients. Patients were randomized to a training group (400 patients，including 133 positive and 267 negative) and a validation group (100 patients，including 41 positive and 59 negative) using a random number method. Using the Faster R-CNN to learn and train 20 430 high-resolution MRI images of the training group，an artificial intelligence image-aided diagnosis platform was established. The 5107 high-resolution MRI images of the validation group were clinically validated. Receiver operating characteristic (ROC) curve and area under the curve (AUC) were used to compare the diagnostic results of the artificial intelligence image-aided diagnosis platform and the senior image expert. Results The accuracy，sensitivity，specificity，positive predictive value and negative predictive value of EMVI for artificial intelligence image-aided diagnosis platform were 93.4%, 97.3%, 89.5%, 0.90 and 0.97，respectively. The area under the receiver operating characteristic curve (AUC) was 0.98. The time required to automatically recognize a single image was 0.2 seconds，which had clear advantages compared to radiologists (estimated to be about 10 seconds). Conclusion    The artificial intelligence image-assisted diagnosis platform based on Faster R-CNN has high efficiency and feasibility for identifying rectal cancer EMVI，and can assist imaging diagnosis.

Key words: artificial intelligence, rectal cancer, extramural vascular invasion, magnetic resonance imaging