【摘要】 目的:构建人工智能(AI)深度学习模型用于辅助CT医生对甲状腺结节进行实时诊断。方法:选取数据库中5218张甲状腺CT图像,其中确诊多发结节图像1200张,单发结节图像1000张,均用于模型训练;其余3018张图像用于AI深度学习模型验证,同时提交给4名CT医师进行诊断、分析和统计。结果:AI深度学习模型对甲状腺图像的正确识别准确率为86.45%,每张图像的诊断时间为(0.12±0.02)s。而两名低年资医师和两名高年资医师的诊断准确率分别为71.4%和77.34%,AI深度学习模型诊断均优于4名医师。结论:构建的AI深度学习模型用于甲状腺结节的诊断具有较高的准确率、特异度和灵敏度,可在CT诊断甲状腺检查中辅助医师进行实时诊断。更多还原

【Abstract】 Objective: To construct a deep learning model with artificial intelligence(AI) so as to assist computed tomography(CT) doctors to implement real-time diagnosis for thyroid nodules. Methods: 5218 CT images of thyroid were selected from the database. In these images, there were 1200 images of confirmed multiple nodules and 1000 images of confirmed single nodules, and they were used in model training. And other 3018 images were used in verification of AI deep learning model, and were submitted to 4 CT doctors for diagnosis, analysis and statistics. Results: The accuracy of corrective recognition of AI deep learning model for thyroid image was 97.6%, and the diagnosis time of each image was(0.12±0.02)s. The accuracies of two junior doctors and two senior doctors were 73.4% and 85.6%, respectively. And AI deep learning model was better than four doctors. Conclusion: The constructed AI deep learning model has high accuracy, specificity and sensitivity when it is used to diagnose thyroid nodules, and it can assist doctors to implement real-time diagnosis in the examination of thyroid nodules by using CT diagnosis.更多还原