基于CT的深度学习模型在甲状腺结节良恶性鉴别中的应用

Study on deep learing model based on CT in the differential diagnosis of benign and malignant thyroid nodules

目的:构建和验证一个用于CT自动识别甲状腺结节的深度学习模型,旨在提高CT医师对甲状腺结节的诊断水平。方法:从医院数据库中选取经手术病理证实的甲状腺结节患者672例,选取330例(其中恶性280例,良性病变50例)用于深度学习模型的训练,余342例用于模型验证,同时再由3名不同年资的CT医师进行诊断,并统计相关结果。结果:深度学习模型用于甲状腺结节的诊断准确率91.8%、敏感度84.5%、特异度87.8%,每例诊断时间为(0.30±0.02)s,均优于3名医师(均P<0.05)。结论:深度学习模型用于甲状腺结节的诊断具有较高的准确率、特异度和敏感度,可辅助CT医师实时诊断甲状腺结节。

Objective:To construct and validate a deep learning model for automatic recognition of thyroid nodules on CT,aiming in improving the level of recognition and diagnosis of thyroid nodules on CT.Methods:672 patients with thyroid nodules confirmed by surgery and pathology were selected from hospital database.330 cases(including 280 malignant cases and 50 benign cases)were selected for the training of the deep learning model.Then the remaining 342 cases were used to validate the model,and three radiologists with different seniority were assigned to diagnose the disease.Finally,the relevant results were counted.Results:The accuracy,sensitivity and specificity of the deep learning model for thyroid nodules were 91.8%,84.5%and 87.8%,respectively.The diagnostic time of each patient was(0.30±0.02)s,which was better than that of the three doctors.Conclusions:The deep learning model constructed in this study has high accuracy,specificity and sensitivity in the diagnosis of thyroid nodules,and can assist CT physicians in real-time diagnosis of thyroid nodules.