| Objective: To evaluate the effectiveness of artificial intelligence systems in the detection of lung nodules and the differentiation of benign and malignant.Methods: A retrospective analysis of the clinical data of 274 patients with pulmonary nodules in the Thoracic Surgery Department of Lanzhou University Second Hospital from May 2016 to December 2020.The preoperative chest CT was imported into two artificial intelligence systems(Scryn ProTM-Small lung nodules intelligent auxiliary screening system V3.3 and Surgi ProTM-Early lung cancer surgery intelligent auxiliary decision-making system V3.0.1),record the diameter,density classification,and malignant risk value of the detected lung nodules.Compare the detection rates of two artificial intelligence systems and radiologists on target lung nodules;compare the sensitivity,specificity,positive likelihood ratio,and negative likelihood of two artificial intelligence systems and radiologists in distinguishing benign and malignant lung nodules Likelihood ratio,draw ROC curve to calculate and compare AUC value,and evaluate the diagnostic efficiency of the artificial intelligence system.Results: A total of 282 pulmonary nodules were diagnosed by surgical resection.The detection rate of the Scryn ProTM system was 100%(282/282),the detection rate of the Surgi ProTM system was 99.65%(281/282),and the detection rate of radiologists was99.65%(281/282).The sensitivity of the Scryn ProTM system for judging benign and malignant lung nodules is 94.70%,the specificity is 25.95%,the positive likelihood ratio is 1.28,and the negative likelihood ratio is 0.20;the sensitivity of the Surgi ProTMsystem for judging benign and malignant lung nodules is 88.00%,Specificity is 57.25%,the positive likelihood ratio is 2.06,the negative likelihood ratio is 0.21;the radiologist's sensitivity to judge benign and malignant pulmonary nodules is 84.40%,specificity is 70.08%,the positive likelihood ratio is 2.82,the negative likelihood ratio is 0.22.The AUC area of the Scryn ProTM system was 0.609(P=0.02(<0.05)),the AUC area of the Surgi ProTM system was 0.723(P=0.00(<0.05)),and the AUC area of radiologists was 0.772(P=0.00(< 0.05)).The performance of the two artificial intelligence systems in distinguishing benign and malignant lung nodules with different sizes.When the lung nodule 5mm?diameter?10mm,the comparison of the three with P=1 indicates that the performance of the three is similar.When pulmonary nodules10mm?diameter?30mm,the Scryn ProTM system has the highest sensitivity at 95.07%,but its specificity is the lowest at 24.41%,and the radiologist's sensitivity is the lowest at 84.21%,but its specificity is the highest,is 69.36%,and the Surgi ProTM system is somewhere in between.The performance of the two artificial intelligence systems in distinguishing benign and malignant in different lung nodule densities classification.For solid nodules,the P=0 of the Scryn ProTM system compared with the Surgi ProTMsystem and the radiologist is reflected in the high sensitivity of the Scryn ProTM system,the specificity is lower than Surgi ProTM system and radiologist;Surgi ProTM system compared with radiologist P=0.082,>0.05,suggesting that Surgi ProTM system and radiologist have similar performance.For subsolid nodules,the P values of the three groups compared with each other were 0.125,0.125,and 0.754,which were all greater than 0.05,indicating that the performance of the three groups was similar.Conclusion: The 2 artificial intelligence pulmonary nodule auxiliary diagnosis systems have a strong performance in the detection of lung nodules,which is equivalent to the level of radiologists;in the identification of benign and malignant pulmonary nodules,with the continuous optimization of the artificial intelligence system,the artificial intelligence system performance is close to the level of radiologists.The artificial intelligence system can be used as an auxiliary tool for radiologists and thoracic surgeon to screen and diagnose lung nodules and guide treatment. |
PDF Full Text Request