Artificial intelligence-driven computer aided diagnosis system provides similar diagnosis value compared with doctors evaluation in lung cancer screening

Abstract

Objective To evaluate the consistency between doctors and Artificial Intelligence (AI) software in analyzing lung nodules, and further compare the advantages of doctors and artificial intelligence in identifying benign and malignant lung nodules based on the pathological results as the gold standard. Materials and methods This retrospective study analyzed participants over 40 years old in the local place from 2011 to 2013. Pulmonary nodules were radiologically examined with a low-dose chest CT scan while being evaluated by a doctor expert panel of radiology, oncology, and thoracic department, and by a computer-aided diagnostic system driven by AI, respectively. Consistency tests were used to evaluate the consistency of the radiological features of the lung nodules. The Receiver Operating Characteristic curve(ROC) was used to evaluate the diagnostic efficacy. Logistic regression analysis was used to predict the possibility of lung cancer. Results Totally 570 subjects were included in this retrospective study. The AI software was very consistent with the panel's evaluation in the position and diameter of the lung nodules (kappa = 0.883, Concordance Correlation Coefficien(CCC) = 0.809, p = 0.000), and the consistency comparison of the solid nodules was acceptable in attenuation characteristics (kappa = 0.503); For patients diagnosed with lung cancer, the AUC of the panel and AI were 0.873 and 0.921, respectively, however, there was no significant difference (p = 0.0950). The diameter is a predictor of lung cancer in both panel and AI predictive models. Conclusion AI software can help doctors diagnose nodules and is consistent with doctors in differentiating between benign and malignant lung nodules.