Large-scale evaluation of multimodal large language models for pneumothorax detection

Key Points

• This research aims to evaluate the effectiveness of multimodal large language models in detecting pneumothorax.
• Conducted a large-scale evaluation of automated systems for pneumothorax detection
• Assessed models for sensitivity and specificity in clinical scenarios
• Analyzed the potential role of models in prioritizing urgent cases
• Automated systems showed high specificity but low sensitivity in pneumothorax detection
• Models may assist in confirming positive findings
• Results indicate the need for careful reliance on automated detection in clinical settings