AI-Generated Medical Reports from Brain CT Scans ?

AI-Generated Medical Reports from Brain CT Scans ?

I’m excited to share a deep learning project I recently completed: an AI-powered system that generates structured medical descriptions from brain CT images. The model combines a ViT (Vision Transformer) with GPT-2 to analyze grayscale CT scans and automatically produce diagnostic-style reports, including:

•? Caption summaries

•? Region of interest localization

•️ Observational notes (e.g., potential hemorrhage)

• JSON-structured outputs ready for integration

️ Key Highlights:

•Trained on real-world labeled data from radiologists

•Used Hugging Face Transformers + PyTorch

•Achieved continuous improvement over epochs with significant loss reduction

•Deployed using an interactive GUI with Gradio for real-time testing

This project demonstrates how AI can assist radiologists by pre-generating reports, accelerating diagnosis, and improving consistency.

Next Steps: I’m exploring how to refine outputs using domain-specific language models and add confidence scoring to each section