全国范围集中学习是一种用于胸腔中部恶性肿瘤诊断的方法：一项多中心队列研究。
Pan-mediastinal neoplasm diagnosis via nationwide federated learning: a multicentre cohort study.
发表日期：2023 Sep
作者：
Ruijie Tang, Hengrui Liang, Yuchen Guo, Zhigang Li, Zhichao Liu, Xu Lin, Zeping Yan, Jun Liu, Xin Xu, Wenlong Shao, Shuben Li, Wenhua Liang, Wei Wang, Fei Cui, Huanghe He, Chao Yang, Long Jiang, Haixuan Wang, Huai Chen, Chenguang Guo, Haipeng Zhang, Zebin Gao, Yuwei He, Xiangru Chen, Lei Zhao, Hong Yu, Jian Hu, Jiangang Zhao, Bin Li, Ci Yin, Wenjie Mao, Wanli Lin, Yujie Xie, Jixian Liu, Xiaoqiang Li, Dingwang Wu, Qinghua Hou, Yongbing Chen, Donglai Chen, Yuhang Xue, Yi Liang, Wenfang Tang, Qi Wang, Encheng Li, Hongxu Liu, Guan Wang, Pingwen Yu, Chun Chen, Bin Zheng, Hao Chen, Zhe Zhang, Lunqing Wang, Ailin Wang, Zongqi Li, Junke Fu, Guangjian Zhang, Jia Zhang, Bohao Liu, Jian Zhao, Boyun Deng, Yongtao Han, Xuefeng Leng, Zhiyu Li, Man Zhang, Changling Liu, Tianhu Wang, Zhilin Luo, Chenglin Yang, Xiaotong Guo, Kai Ma, Lixu Wang, Wenjun Jiang, Xu Han, Qing Wang, Kun Qiao, Zhaohua Xia, Shuo Zheng, Chenyang Xu, Jidong Peng, Shilong Wu, Zhifeng Zhang, Haoda Huang, Dazhi Pang, Qiao Liu, Jinglong Li, Xueru Ding, Xiang Liu, Liucheng Zhong, Yutong Lu, Feng Xu, Qionghai Dai, Jianxing He
来源：
Brain Structure & Function
摘要：
纵隔肿瘤是典型的胸部疾病，在全球普通人群中发病率呈上升趋势，并可能导致预后不良。在临床实践中，纵隔的复杂解剖结构和不同纵隔肿瘤病理类型之间的混淆严重阻碍了准确诊断。为解决这些困难，我们基于隐私保护的联邦学习，组织了一个多中心国家合作，并开发了基于胸部CT的人工智能（AI）纵隔肿瘤诊断系统CAIMEN。在这个多中心队列研究中，我们从中国的24个中心收集了7825例纵隔肿瘤病例和796个正常对照组，以开发CAIMEN。我们进一步采用了多视图、知识传递和多级决策模式，对CAIMEN进行了多项新算法的增强。通过内部测试集（15个中心的929例病例）、外部测试集（5个中心的1216例病例和11162例真实世界队列）以及人工智能-人类测试集（来自4个中心的60例阳性病例和15个机构的放射科医师）来评估CAIMEN的检测、分割和分类性能。在外部测试实验中，CAIMEN检测纵隔肿瘤的受试者工作特征曲线下面积为0.973（95％CI 0.969-0.977）。在真实世界队列中，CAIMEN检测到了经放射科医师确认的13例假阴性病例。CAIMEN对于分割纵隔肿瘤的Dice分数为0.765（0.738-0.792）。CAIMEN的纵隔肿瘤分类top-1和top-3的准确率分别为0.523（0.497-0.554）和0.799（0.778-0.822）。在人工智能-人类测试实验中，CAIMEN在top-1和top-3准确性方面优于临床医生，分别为0.500（0.383-0.633）和0.800（0.700-0.900）。与此同时，在基于CAIMEN的辅助计算机辅助诊断软件的帮助下，46名临床医生的平均top-1准确率提高了19.1％（0.345-0.411），top-3准确率提高了13.0％（0.545-0.616）。对于纵隔肿瘤，CAIMEN能够产生高诊断准确性并协助人类专家的诊断，展示了其在临床实践中的潜力。中国国家重点研发计划、国家自然科学基金和北京自然科学基金资助。© 2023 作者。由 Elsevier Ltd. 发表。本文是根据CC BY-NC-ND 4.0许可证的开放获取文章。© 由 Elsevier Ltd. 发表。保留所有权利。
Mediastinal neoplasms are typical thoracic diseases with increasing incidence in the general global population and can lead to poor prognosis. In clinical practice, the mediastinum's complex anatomic structures and intertype confusion among different mediastinal neoplasm pathologies severely hinder accurate diagnosis. To solve these difficulties, we organised a multicentre national collaboration on the basis of privacy-secured federated learning and developed CAIMEN, an efficient chest CT-based artificial intelligence (AI) mediastinal neoplasm diagnosis system.In this multicentre cohort study, 7825 mediastinal neoplasm cases and 796 normal controls were collected from 24 centres in China to develop CAIMEN. We further enhanced CAIMEN with several novel algorithms in a multiview, knowledge-transferred, multilevel decision-making pattern. CAIMEN was tested by internal (929 cases at 15 centres), external (1216 cases at five centres and a real-world cohort of 11 162 cases), and human-AI (60 positive cases from four centres and radiologists from 15 institutions) test sets to evaluate its detection, segmentation, and classification performance.In the external test experiments, the area under the receiver operating characteristic curve for detecting mediastinal neoplasms of CAIMEN was 0·973 (95% CI 0·969-0·977). In the real-world cohort, CAIMEN detected 13 false-negative cases confirmed by radiologists. The dice score for segmenting mediastinal neoplasms of CAIMEN was 0·765 (0·738-0·792). The mediastinal neoplasm classification top-1 and top-3 accuracy of CAIMEN were 0·523 (0·497-0·554) and 0·799 (0·778-0·822), respectively. In the human-AI test experiments, CAIMEN outperformed clinicians with top-1 and top-3 accuracy of 0·500 (0·383-0·633) and 0·800 (0·700-0·900), respectively. Meanwhile, with assistance from the computer aided diagnosis software based on CAIMEN, the 46 clinicians improved their average top-1 accuracy by 19·1% (0·345-0·411) and top-3 accuracy by 13·0% (0·545-0·616).For mediastinal neoplasms, CAIMEN can produce high diagnostic accuracy and assist the diagnosis of human experts, showing its potential for clinical practice.National Key R&D Program of China, National Natural Science Foundation of China, and Beijing Natural Science Foundation.Copyright © 2023 The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY-NC-ND 4.0 license. Published by Elsevier Ltd.. All rights reserved.