[Establishment and clinical testing of pancreatic cancer Faster R-CNN AI system based on fast regional convolutional neural network].

• DOI: 10.3760/cma.j.cn112139-20191017-00515
• Publication Date: 2020-07-01
• Authors: Sen Yang; Yun Lu; Xiuqi Zheng; Yi‐xin Zhang; F J Xin; Peng Sun; Yu Li; Shuo Liu; Shu Li; Yinan Guo; Shaocheng Liu
• Keywords: Pancreatic carcinoma

Objective: To investigate the effectiveness of an enhanced CT automatic recognition system based on Faster R-CNN for pancreatic cancer and its clinical value. Methods: In this study, 4 024 imaging sequences 315 patients with from January 2013 to May 2016 at Affiliated Hospital Qingdao University were collected retrospectively, 2 614 input into faster as training dataset create image model, which was then validated by reading 1 410 images 135 cases cancer.In order identify effectiveness, 3 750 150 lesions read a followed-up carried out.The accuracy recall rate in detecting nodules recorded regression curves generated.In addition, accuracy, sensitivity specificity diagnosis analyzed, ROC generated area under calculated. Results: Based cases, curve 0.927 calculated R-CNN. The 0.902, 0.913 0.801 respectively.After data verified, 893 showed positive 857 negative.Ninety-eight diagnosed R-CNN.After follow-up, it found that 53 post-operatively proved be ductal carcinoma, 21 cystadenocarcinoma, 12 cystadenoma, 5 cyst, 7 untreated.During 17 months after operation, 6 died abdominal tumor infiltration, liver lung metastasis.Of 52 who negative R-CNN, 9 carcinoma. Conclusion: has value helping physicians diagnose cancer.目的： 验证基于快速区域卷积神经网络（Faster R-CNN）胰腺癌增强CT自动识别系统，并探讨其临床应用价值。 方法： 回顾性收集青岛大学附属医院2013年1月至2016年5月收治的315例胰腺癌患者的4 024张增强CT影像序列，将2 614张影像序列作为训练组输入Faster R-CNN系统，建立影像自动识别模型，通过读取135例胰腺癌的1 410张增强CT影像进行验证。为了进一步测试其临床应用效果，读取150例胰腺占位患者的3 750张增强CT影像并对其诊断结果进行随访。记录结节类别的精准率和召回率，绘制精确回归曲线，分析Faster R-CNN诊断的准确性、灵敏度、特异度，生成受试者工作特征（ROC）曲线，并计算曲线下面积。 结果： 基于135例胰腺癌增强CT影像，得到Faster R-CNN的人工智能辅助诊断的ROC曲线的曲线下面积为0.927，准确性、特异度、灵敏度分别为0.902、0.913、0.801。经过150例胰腺占位患者资料的验证，判定阳性893张，阴性2 857张，Faster R-CNN诊断为胰腺癌患者98例，对其诊断结果进行随访，其中53例经外科手术后病理证实为胰腺导管癌、21例为胰腺囊腺癌、12例为胰腺囊腺瘤、5例为胰腺囊肿，7例患者未手术治疗。在术后5~17个月内6例死于腹腔肿瘤浸润、肝转移或肺转移。在Faster R-CNN诊断为阴性的52例患者中，有9例经外科术后证实为胰腺导管癌。 结论： R-CNN系统能够帮助影像科医师对胰腺癌进行诊断，具有一定的临床应用价值。.