AI-driven nodule detection in chest X-rays: Validation with radiologist-confirmed CT and X-ray findings

Objectives: This study evaluates the performance of a commercially available deep learning tool in detecting chest nodules on X-rays, with findings validated by radiologist and confirmatory Computed Tomography (CT) scans. Methods: In this retrospective analysis, the data of 299 consecutive patients who underwent both chest X-rays and CT scans within two weeks from June 2024 to December 2024 were analyzed. The performance of the deep learning tool was compared against the radiologist reports and CT scan reports, which were considered the gold standard. Performance parameters such as accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated and compared using McNemar's test. Results: In 299 patients (43.8% female, mean age 58.78 years), CT scans showed nodules in 122 (40.8%) patients. Nodules were identified by the deep learning system in 53 patients, missing five cases identified by radiologist, but also identifying an additional 37 (30.3%) missed cases. The deep learning tool was 72% accurate (95% confidence interval: 0.66-0.77), higher than the 63% of the radiologist (P=0.02). Sensitivity was higher for artificial intelligence (37%) than for radiologist (10%; P<0.001) but lower in terms of specificity (95% vs. 100%; P=0.004). Conclusions: The deep learning algorithm showed improved sensitivity and accuracy for the detection of pulmonary nodules but lower specificity, with reservations about false positives. The synergy of artificial intelligence and human experience has the potential to enhance the diagnosis of lung cancer. Further research is needed to validate these findings in diverse populations.