The Potential Role of GPT-4o in Interpreting Chest Computed Tomography for Thoracic Emergencies

This study evaluated GPT-4o's performance in interpreting chest CT images for thoracic emergencies. In July 2024, open-access medical images were analyzed by GPT-4o via chatgpt.com with a query to identify and describe abnormalities, specifying the type and plane of imaging. GPT-4o consistently identified the imaging type and plane correctly. For instance, it accurately detected a right-sided pneumothorax with mediastinal shift, and a significant right pleural effusion with a smaller left effusion, although it misidentified minimal bilateral pneumothorax. It correctly identified pericardial effusion in the coronal plane in a cardiac tamponade case but failed to identify a saddle pulmonary embolism. While GPT-4o shows potential for diagnosing chest CT images in thoracic emergencies, it may still miss major abnormalities. Future large-scale studies should evaluate both common and rare pathologies using large language models.