基于深度学习的口腔鳞状细胞癌CT增强图像中颈部转移淋巴结自动检测模型的建立及临床应用研究  

Establishment and clinical application of automatic detection model of cervical lymph node metastasis in contrast-enhanced CT images of oral squamous cell carcinoma based on deep learning

作　　者：苏潇[1] 吴超逸[2] 常峰 巫智涵 孙嘉诚 林绩腾 周宏润 陶晓峰[1] 朱凌[1] SU Xiao;WU Chaoyi;CHANG Feng;WU Zhihan;SUN Jiacheng;LIN Jiteng;ZHOU Hongrun;TAO Xiaofeng;ZHU Ling(Department of Radiology,Shanghai Ninth People’s Hospital,Shanghai Jiao Tong University School of Medicine,Shanghai 200011,China;Cooperative Medianet Innovation Center,Shanghai Jiao Tong University,Shanghai 200240,China;College of Stomatology,Shanghai Jiao Tong University,Shanghai 200025,China)

机构地区：[1]上海交通大学医学院附属第九人民医院放射科,上海200011 [2]上海交通大学未来媒体网络协同创新中心,上海200240 [3]上海交通大学口腔医学院,上海200025

出　　处：《肿瘤影像学》2023年第1期20-25,共6页Oncoradiology

摘　　要：目的:基于深度学习,建立一个在颌面颈部增强计算机体层成像(computed tomography,CT)图像上,完成口腔鳞状细胞癌(oral squamous cell carcinoma,OSCC)患者颈部转移淋巴结自动检出的模型。方法:收集114例OSCC患者的颌面颈部增强CT扫描图像,所有勾勒的转移淋巴结均得到病理学检查证实(共216枚),图像层厚为0.625 mm,单层图像分辨率512×512。随机分为训练集80例,测试集34例。以上结果经过深度学习模型的训练和验证,评估其自动检测转移淋巴结的可行性。结果:转移淋巴结自动检测模型的自由响应受试者工作特征(free-response receiver operating characteristic,FROC)@1:0.3915;FROC@2:0.5183;FROC@3:0.6478;FROC@4:0.7408;FROC@5:0.8169;FROC@6:0.8535;mFROC:0.6615;maxF1-score:0.4385;灵敏度的最佳表现为87.32%。结论:本研究建立的深度学习模型可用于颌面颈部增强CT图像中的OSCC患者颈部转移淋巴结的自动检测,为OSCC患者转移淋巴结快速自动检测提供了新方法,有利于实现头颈影像专科医师知识的下沉及提高初级影像科医师的培训效率。Objective:To establish a deep-learning model for automatically detecting metastatic lymph nodes(LN)of oral squamous cell carcinoma(OSCC)patients from contrast-enhanced computed tomography(CT)images.Methods:Contrast-enhanced CT images of 114 oral cancer patients were collected.The metastatic LNs of these patients,a total of 216,had been pathologically conflrmed.All CT scans are with a slice thickness of 0.625 mm and resolution is 512×512.It was randomly divided into a training set of 80 cases and a test set of 34 cases.The above results were trained and verifled by a deep learning model.Performance in detecting metastasis were obtained.Results:Performance in detecting metastatic LNs showed FROC@1 of 0.3915,FROC@2 of 0.5183,FROC@3 of 0.6478,FROC@4 of 0.7408,FROC@5 of 0.8169,FROC@6 of 0.8535,mFROC of 0.6615,maxF1-score of 0.4385,the best performance of sensitivity is 87.32%.Conclusion:A deep-learning model can be used to automatically detect metastatic LNs in contrast-enhanced CT images of patients with OSCC,which provides a new idea for the rapid detection of metastatic LNs and realize the spread of knowledge of radiologists of head and neck imaging and improve the training efficiency of primary radiologists.

关 键 词：口腔鳞状细胞癌 计算机体层成像 淋巴结转移 淋巴结分区 深度学习 

分 类 号：R739.8[医药卫生—肿瘤] R445.3[医药卫生—临床医学]