To evaluate the value of risk assessment of artificial intelligence (AI) in the differential diagnosis of benign and malignant pulmonary nodules.

All of 310 patients with pulmonary nodules by the Chest CT examination in Tangdu Hospital from August 2018 to December 2019 were collected in this study. A copy of the patient′s CT image with DICOM format was input into the "FACT artificial intelligence" software system to and the pulmonary nodules was analyzed. The pulmonary nodules characteristics, including the location, quantity, featutes (ground glass, subsolid and solid), size, density and AI values and Lung-rads grade were obtained. After multidisciplinary discussion, 39 cases of pulmonary nodules were suggested to be diagnosed by surgery, percutaneous lung puncture or bronchoscopic biopsy and 271 patients were followed up.

Among 31 cases of pulmonary nodules, 14 cases were benign, including tuberculosis (8 cases), cryptococci (2 cases), inflammatory nodule (4 cases), and 25 cases were malignant, including squamous cell carcinoma (2 cases) and adenocarcinoma (23 cases). Further analysis showed that the AI risk probability of malignant lesions was significantly higher than that of benign lesions (P>0.05), and the AI risk probability of nodules was significantly correlated with the characteristics of pulmonary nodules (ground glass, subsolid and solid) (P>0.05), but not with the numbers and the marginal burr sign (P>0.05). There were significant differences in the characteristics of pulmonary nodules (ground glass, subsolid, and solid) between benign and malignant (P<0.05), but there was no significant difference in density or volume between benign and malignant (P>0.05). In addition, Lung-rads grade significantly correlated with AI risk probability of pulmonary nodules (P<0.05).

The automatic analysis of benign and malignant probability of pulmonary based on AI has a certain value in the differential diagnosis of pulmonary nodules and could be used in clinic.