The introduction of low-dose computed tomography (LDCT) for lung cancer screening has faced the problem of a high rate of false positives, leading to unnecessary invasive procedures and an increased burden on the healthcare system. Artificial intelligence systems have the potential to optimize this process, but their actual effectiveness in clinical practice needs to be studied. A retrospective study of 100 LDCT studies with a verified diagnosis was conducted. Four radiologists (an expert and three residents) independently analyzed the studies without using the Philips IntelliSpace Discovery 3.0 AI system. Sensitivity, specificity, frequency of false positive results, analysis time, and interoperative consistency were evaluated. The use of an AI assistant significantly improved the performance of residents: sensitivity increased by 11.4-14.3%, and the frequency of false positive results decreased by 6.1–10.7%. The analysis time was reduced by 26.6–35.1% for all specialists. Interoperation consistency increased from 0.62 to 0.81. No significant changes have been identified for the expert. There was a significant reduction in false-positive conclusions of categories 4A and 4B on the Lung-RADS scale. The integration of artificial intelligence into lung LDCT analysis significantly improves the efficiency of novice specialists, reduces diagnostic time, and reduces the frequency of clinically significant false positive results, which is especially important for mass screening.
