The role of AI assistance in the evaluation of unenhanced chest CT scans in an emergency setting

Purpose: To evaluate the potential role of artificial intelligence (AI)-based software in assisting radiologists with reporting unenhanced chest computed tomography (CCT) scans in an emergency setting. Material and methods: It was an IRB-approved retrospective study, and the need for informed consent was waived. We included 90 unenhanced CCT scans performed in an emergency setting over a 2-month period (November-December 2024). Anonymized original reports were retrieved. Axial 3 mm thick multiplanar reconstructions were processed using commercially AI-based software (xAid Chest, xAID LLC, Barcelona, Spain). All scans were subsequently re-evaluated by two radiologists in consensus (reference standard). Detection of lung nodules, lung opacifications, emphysema, coronary calcification, aortic dilatation, pulmonary dilatation, pleural effusion, pericardial effusion, pneumothorax, rib fractures, vertebral fractures, and adrenal masses was compared between original reports, AI outputs, and image revision. Results: In the original reports, the frequency of reported findings ranged from 96.7% (pleural effusion) to 5.6% (pulmonary artery dilatation); among the described findings, the positivity rate ranged from 100% (emphysema) to 11.4% (pericardial effusion). The AI software demonstrated non-inferior sensitivity and specificity compared to the reporting radiologist in terms of sensitivity in all pathologies, excluding emphysema. For several findings that are not routinely reported by radiologists (coronary calcifications, pulmonary dilatation, vertebral fractures), the AI system outperformed the radiologist in sensitivity, albeit with a trade-off in specificity. Conclusion: AI is a valuable tool for assisting radiologists in reporting unenhanced CCT scans in an emergency setting.