Validation of Artificial Intelligence in Detection of Indeterminate Pulmonary Nodules in Adults by Chest Radiography

Psoriatic Lung cancer remains the leading cause of cancer-related mortality worldwide. Early detection of indeterminate pulmonary nodules (IPNs) is critical for improving patient outcomes, as timely intervention correlates with higher survival rates. Aim To evaluate the diagnostic performance of the qXR deep-learning algorithm for detecting pulmonary nodules or masses on standard chest radiographs (CXR), using follow-up computed tomography (CT) as the reference standard, and to characterize confirmed nodules according to the Fleischner Society 2017 guidelines. In this single-center, retrospective cohort study, 6,543 CXRs obtained at Baghdad Teaching Hospital between January and December 2023 were screened by the qXR algorithm.  Of these, 1,320 images were flagged positive for nodular or mass-like opacities.  After applying exclusion criteria—prior thoracic surgery, poor image quality, refusal of follow-up, or nodules < 6 mm—a total of 83 cases underwent CT within two weeks.  Algorithm performance was assessed on a per-scan and per-nodule basis, with CT findings serving as the reference.  CT confirmed pulmonary nodules in 65 of 83 cases, yielding a per-scan positive predictive value (PPV) of 78.4%.  The qXR algorithm detected 116 nodules on CXR versus 94 nodules identified by CT; 73.4% of confirmed nodules were solid. For solid nodules > 8 mm, sensitivity reached 92.5%. Overall, the area under the receiver-operating characteristic curve (AUC) was 0.68, with scan-level sensitivity and specificity of 92.5% and 81.1%, respectively. Subgroup analyses demonstrated variations in performance according to patient demographics (age, sex, smoking history) and nodule characteristics (size, solidity, lobe location). The qXR algorithm exhibits high sensitivity and acceptable specificity for detecting indeterminate pulmonary nodules on CXR, particularly for solid lesions > 8 mm. Integration of AI-driven detection into routine chest radiograph interpretation has the potential to enhance early nodule identification, expedite clinical decision-making, and support lung-cancer risk stratification without replacing radiologists.