baichuan-inc/BaichuanMed-OCR-7B

BaichuanMed-OCR-7B: Medical Report OCR and Q&A

BaichuanMed-OCR-7B is a 7 billion parameter model developed by baichuan-inc, fine-tuned from Qwen2.5-VL-7B-Instruct. It has been specifically adapted using curated medical report datasets, comprising images and associated questions and answers, to address the unique challenges of medical document processing.

Key Capabilities

• Robust Medical OCR: Accurately extracts complex textual information, including tables and key symbols, from medical report images.
• Structured Output: Supports generating extracted information in a structured Markdown format, facilitating downstream processing and analysis.
• Accurate Comprehension and Generation: Understands user queries and produces relevant, logically sound answers based on the text identified through OCR.

Performance Highlights

Evaluated on a private medical report dataset, BaichuanMed-OCR-7B achieved an accuracy of 83.5%, significantly outperforming its base model, Qwen2.5-VL-7B-Instruct (71.3%). This demonstrates its strong specialization and effectiveness in tasks such as extracting critical information and summarizing content from medical reports.

Ideal Use Cases

• Automated Medical Data Entry: Streamlining the digitization of information from scanned medical reports.
• Clinical Decision Support: Assisting healthcare professionals by quickly extracting and summarizing relevant patient data from image-based reports.
• Medical Research: Facilitating the analysis of large volumes of medical report data by converting unstructured image content into queryable text.