Towards Automated Reporting: A Bronchoscopy Report Dataset for Enhancing Multimodality Large Language Models

Bronchoscopy examination is essential for diagnosing and managing respiratory diseases. While Multimodality Large Language Models (MLLMs) can enhance the efficiency and accuracy of medical report writing, existing datasets lack descriptive and comprehensive annotations for complex cases, hindering their ability to facilitate adequate learning of image-report relationships. To address this problem, we introduce BERD, a Bronchoscopy Examination Report Dataset, which includes 3,692 bronchoscopy examination reports. Among these reports, 6,330 representative images are annotated with single-image text descriptions and classification labels. BERD emphasizes the provision of versatile and detailed descriptions of findings. All these reports and annotations were performed by experienced clinicians specializing in bronchoscopy. Furthermore, experimental results show that fine-tuning state-of-the-art MLLMs on BERD significantly improves their ability to generate accurate and comprehensive reports, advancing AI applications in bronchoscopy.