| Objectives:(1)To select an artificial intelligence model based on electronic computed tomography images for the diagnosis of lateral cervical lymph nodes in patients with papillary thyroid cancer.(2)To analyze the results of the selected AI model for diagnosing the presence of metastasis in the lateral cervical lymph nodes of patients with papillary thyroid cancer and compare them with the results given by physicians specializing in imaging to judge the diagnostic performance of the AI model and its application value in clinical work.Methods:(1)Image acquisition:CT images of 196 patients who had been pathologically diagnosed as papillary thyroid cancer were selected from the database of the Second Affiliated Hospital of Zhejiang University,including 71 male patients and 125 female patients,whose ages ranged from 15 to 68 years old,with an average age of(40.5± 12.5)years,and 676 cases of lymph nodes in the lateral cervical region were counted,including 103 cases of benign lymph nodes and 573 cases of malignant lymph nodes confirmed by pathology.(2)Pre-processing of the images:The lateral cervical lymph nodes with clear diagnosis and complete structure in all CT images were marked by an imaging physician with extensive experience in the processing of cervical computed tomography.Each labeled lymph node was set as a separate jpg file,and the work was reviewed by another experienced imaging physician after completion to ensure the accuracy of the work.(3)Artificial intelligence selection and training:By comparing the different artificial intelligence models which are currently used in clinical practice,the model with higher application value was selected,and 473 lymph node images(72 non-metastatic lymph nodes and 401 metastatic lymph nodes)were selected through random sampling,and used for artificial intelligence deep learning model training.(4)AI deep learning model validation:203 lymph node CT images(31 non-metastatic lymph nodes and 172 metastatic lymph nodes)remaining in the sample were used for AI deep learning model validation,and their accuracy,sensitivity and specificity were counted.(5)Manual interpretation by imaging professionals:The 203 lymph node computed tomography images used for AI deep learning model validation were submitted to a senior imaging professional and two junior imaging physicians for diagnosis and analysis.(6)Statistics and analysis:The diagnoses made by the AI model,the diagnoses made by the senior imaging specialist,and the diagnoses made by the junior imaging specialist were statistically analyzed.The results of the diagnoses made by the AI deep learning model and the diagnoses made by physicians with different experience in imaging were compared and analyzed.Result:The accuracy of the AI deep learning model in correctly identifying lymph node metastasis in the cervical region was 0.926(188/203),sensitivity was 0.948(163/172),and specificity was 0.806(25/31);the diagnostic accuracy of the senior imaging specialist was 0.783(159/203),sensitivity was 0.791(136/172),and specificity was 0.742(23/3 1);the diagnostic accuracy of junior imaging specialist 1 was0.749(152/203),sensitivity was 0.762(131/172),specificity was 0.677(21/31);and the diagnostic accuracy of junior imaging specialist 1 was 0.759(154/203),sensitivity was 0.791(136/172),specificity was 0.581(18/31).Conclusion:The AI deep learning model constructed for the diagnosis of lymph node metastasis in the lateral cervical region of papillary thyroid cancer patients has higher accuracy,specificity and sensitivity compared with the imaging physicians,which can effectively improve the accuracy of diagnosis and provide better imaging support for clinical diagnosis. |
PDF Full Text Request