# 🩻 Chest X-ray Report Generator using Clinical-T5 and RAD-DINO

A fully functional, medically formatted PDF report generator for chest X-rays. Built using Streamlit, this system combines image embeddings from **RAD-DINO** with a **fine-tuned Clinical-T5** model to generate clean, structured radiology findings. It includes a professional letterhead, clinic logo, digital signature, and stores logs of all reports.

---

## 🧠 Key Features

- 📤 Upload chest X-ray images (`.jpg`, `.png`)
- 🤖 Generate bullet-pointed **radiology findings** using RAD-DINO + Clinical-T5
- 📄 Auto-generate multi-page PDF reports with:
  - Clinic header
  - Patient details
  - Structured findings
  - Doctor signature & clinic logo
  - X-ray image on the second page
- 📊 Log every report to `report_logs.csv`
- 💬 Optional extension for:
  - Sending email with report
  - QR-code based report verification

---

## 🗂 Project Structure

chestxray_report_app/ ├── app.py # Main Streamlit app ├── requirements.txt # Python dependencies ├── README.md ├── .gitignore ├── models/ │ └── clinical_t5_final/ # Pretrained Clinical-T5 model ├── utils/ │ └── dino_embedding.py # RAD-DINO feature extraction ├── assets/ │ ├── logo.png # Clinic logo │ └── signature.png # Doctor’s digital signature └── outputs/ ├── reports/ # Generated PDFs & images └── report_logs.csv # Logged metadata

---

## 🚀 Setup Instructions

### 📥 Clone the Repo

```bash
git clone https://github.com/yourusername/chestxray-report-generator.git
cd chestxray-report-generator

🛠️ Create Environment & Install Requirements

conda create -n cxr-report-gen python=3.10 -y
conda activate cxr-report-gen
pip install -r requirements.txt

---

▶️ Running the App

streamlit run app.py

App will launch at: http://localhost:8501

---

📄 Sample Output

• PDF Includes:
  • Patient metadata
  • Clean, grammatically formatted findings (auto-corrected)
  • X-ray image (page 2)
  • Header/footer with clinic info
  • Doctor's name & digital signature
• CSV Log:
  • Stored in outputs/report_logs.csv
  • Includes patient name, dates, report path

---

🧠 Model Info

• Image Encoder: StanfordAIMI/RAD-DINO
• Text Generator: Fine-tuned Clinical-T5-Sci
• Prompt Engineering: Uses semantic token prompts derived from image features

---

🔐 Customization

Feature	File	Notes
Clinic Logo	assets/logo.png	Displayed top-left of report
Doctor Signature	assets/signature.png	Displayed bottom-right on all pages
T5 Model	models/clinical_t5_final/	Replace with any fine-tuned model
Report Format	create_letterhead_pdf() in app.py	Modify layout or styles

---

🧱 Future Enhancements

• ✅ Email report directly via SMTP
• ✅ QR code on PDF for online verification
• 🧾 Impression section generation (next step)
• 📡 PACS/DICOM or FHIR integration

---

📜 License

This project is licensed under the MIT License.

---

👨‍⚕️ Author

Yash Verma
Clinical-AI & Multimodal Fusion Researcher
NLP_Project | 2025

---

🤝 Contributions

Feel free to fork this repo and submit pull requests!
Want help deploying or extending? Open an issue or reach out.