A Study On Detection Ability Of Deep Learning Technology For Benign And Malignant Pulmonary Nodules Of Chest CT

Object:With the continuous development of deep learning technology and artificial neural networks,artificial intelligence has played an increasingly important role in the diagnosis of diseases by medical imaging.Compare the detection results of benign and malignant lung nodules based on deep learning technology with the diagnostic results of radiation nodule physicians,evaluate the two and the ability of early diagnosis of lung cancer after combination,and improve the resident nodule good The malignant identification ability provides a reference direction for the development of benign and malignant lung nodule screening systems,and improves the screening ability of early lung cancer in the hope that it can provide patients with an early and accurate diagnosis of benign and malignant lung nodules,and prolong it.Methods:In this study,we compiled all 3494 patients with clear pathological results from fiberoptic bronchoscopy,surgery or percutaneous lung biopsy from the First Affiliated Hospital of Medical College of Shihezi University from January 2015 to December 2018.There were 997 patients diagnosed with lung cancer.A return visit was performed to collect chest CT examinations of all patients with malignant nodules from January 2013 to the time of pathological diagnosis.A total of 287 malignant nodules were screened as CT tests with lung nodules at the same location as the tumor.All 287 nodule DICOM standard format images were sorted and imported into the artificial intelligence screening system,and the size,location and malignant probability of the nodule detected by the model were recorded.At the same time,the CT number of the examination number(No.A)corresponding to the 287 pulmonary nodule cases was retrieved by the PACS system,and 2 residents who were participating in the standardized training of residents were read through the blind method to determine the benign and malignant lung nodules in the test set..The number of detected nodules,the number of true positives,and the number of false negatives after the diagnosis of artificial intelligence and the nodule of each resident and the combination of the artificial intelligence screening system and the diagnosis of the resident were recorded separately.SPSS20.0 was used for statistical analysis of all the data,and the sensitivity,misdiagnosis rate,and missed diagnosis rate of nodules detected in each group were compared by tests.P<0.05 was considered statistically significant.Result:The detection rate of malignant pulmonary nodules by the deep learning technology system based on artificial intelligence is 97.21%,which is significantly higher than the 83.97%detection rate of the residents who are participating in the standardi zed training of residents(?~2=29.518,P<0.01).The sensitivity of resident doctors to diagnose malignant pulmonary nodules was 88.82%,which was higher than 66.67%of that of artificial intelligence(?~2=5.552,P<0.05).The detection rate was 98.26%and the sensitivity was 82.62%.Through separate comparative analysis of solid nodules and ground glass nodules,the difference of positive detection rate,missed diagnosis rate and misdiagnosis rate in nodules diagnosis between combined diagnosis and resident diagnosi s alone was statistically significant(P<0.05).Conclusion:1.The deep learning-based artificial intelligence syst em has the same diagnostic capability for different types of pulmonary nodules.2.Compared with the single diagnosis,the imaging resident a ssisted by artificial intelligence has an obvious advantage in the diagnosis of benign and malignant lung nodules i n the early stage of lung cancer,and can reduce the missed diagnosis rate of lung nodules in the early stage of lung cancer,especially in the diagnosis of ashy nodu les.