Leveraging Artificial Intelligence in Radiology: Enhancing the Detection of Subtle Pulmonary Nodules with Deep Learning Algorithms Introduction

Background: Early detection of pulmonary nodules is crucial for lung cancer outcomes, but subtle nodules often go undetected in routine radiological evaluations. This study evaluates a deep learning algorithm's effectiveness in detecting subtle pulmonary nodules on chest CT scans compared to radiologist interpretation. Material and Methods: In this retrospective study, we analyzed 1,200 chest CT scans containing pulmonary nodules ≤8mm. A deep learning algorithm was developed and compared against junior (<5 years) and senior (>10 years) radiologists. Performance metrics included sensitivity, specificity, diagnostic accuracy, and time efficiency. Findings The AI model achieved 91.7% sensitivity and 88.3% specificity (AUC-ROC: 0.93), outperforming junior radiologists (sensitivity: 83.3%, specificity: 85.0%) and matching senior radiologists (sensitivity: 89.2%, specificity: 86.7%). Average interpretation time was significantly reduced with AI (1.2 minutes) compared to junior (4.8 minutes) and senior radiologists (3.5 minutes). However, the AI model showed decreased performance with very low-contrast nodules (sensitivity: 72.5%) compared to senior radiologists (sensitivity: 81.4%). The deep learning algorithm demonstrates promising results in pulmonary nodule detection, particularly in reducing diagnostic time and variability. While it shows comparable performance to experienced radiologists in most scenarios, challenges remain in detecting very low-contrast nodules, suggesting its optimal use as a complementary tool rather than a replacement for radiologist expertise.