LLaVA-Med v1.5 Mistral for Chest X-Ray Analysis

Paper on arXiv: Enhancing Cognition and Explainability of Multimodal Foundation Models with Self-Synthesized Data

This model is a fine-tuned multimodal foundation model based on LLaVA-Med v1.5 Mistral-7B, optimized for analyzing chest X-ray images and detecting pneumonia using the Chest X-Ray Images (Pneumonia) dataset from Kaggle.

Key Details

Base Model: LLaVA-Med v1.5 Mistral-7B
Dataset: Chest X-Ray Images (Pneumonia)
Innovation:
- Self-Synthesized Data: Enhances interpretability by generating human-understandable diagnostic insights.
- Domain-Specific Fine-Tuning: Optimized on medical imaging for accurate pneumonia classification.
- Iterative Training: Utilizes rejection sampling to improve diagnostic accuracy and explanation quality.
Intended Use: Assisting in pneumonia diagnosis from chest X-ray images with detailed, explainable outputs.

How to Use

import requests
from PIL import Image
import torch
from transformers import AutoProcessor, LlavaForConditionalGeneration

model_id = "YuchengShi/llava-med-v1.5-mistral-7b-chest-xray"
model = LlavaForConditionalGeneration.from_pretrained(
    model_id, 
    torch_dtype=torch.float16, 
    low_cpu_mem_usage=True,
).to("cuda")
processor = AutoProcessor.from_pretrained(model_id)

conversation = [
    {
      "role": "user",
      "content": [
          {"type": "text", "text": "Can you analyze this chest X-ray?"},
          {"type": "image"},
        ],
    },
]
prompt = processor.apply_chat_template(conversation, add_generation_prompt=True)
image_file = "chest-xray/test1.png"
raw_image = Image.open(requests.get(image_file, stream=True).raw)
inputs = processor(images=raw_image, text=prompt, return_tensors='pt').to("cuda", torch.float16)

output = model.generate(**inputs, max_new_tokens=200, do_sample=False)
print(processor.decode(output[0][2:], skip_special_tokens=True))

Training & Evaluation

Training: Fine-tuned using LoRA on Chest X-ray images (Pneumonia dataset) with iterative rejection sampling.
Evaluation: Achieves robust pneumonia classification with interpretable diagnostic explanations.

Citation

If you use this model, please cite:

@inproceedings{
  shi2025enhancing,
  title={Enhancing Cognition and Explainability of Multimodal Foundation Models with Self-Synthesized Data},
  author={Yucheng Shi and Quanzheng Li and Jin Sun and Xiang Li and Ninghao Liu},
  booktitle={The Thirteenth International Conference on Learning Representations},
  year={2025},
  url={https://openreview.net/forum?id=lHbLpwbEyt}
}