CultriX's picture
Initial commit without images
3df4685
---
title: MedGemma Radiology Report Generator
emoji: 🩻
colorFrom: purple
colorTo: indigo
sdk: gradio
python_version: '3.10'
app_file: app.py
sdk_version: 5.39.0
---
# 🏥 MedGemma Radiology Report Generator
### Created by **CultriX**
This Hugging Face Space demonstrates the capabilities of Google's **MedGemma 4b-it** model, a medical-focused LLM designed for both image comprehension and medical text generation.
The application is accelerated by **ZeroGPU** for fast inference and features a dual-tab interface for different clinical workflows.
---
### ✨ Features
1. **🩻 X-Ray Analysis Tab**
* **Multimodal Analysis:** Upload an X-ray to receive a structured radiology report (Findings, Impression, Recommendations).
* **Clinical Context:** Optionally provide patient history (e.g., "65M, cough for 3 weeks") to guide the model's interpretation.
* **Interactive Chat:** Ask follow-up questions about specific findings directly in the chat window after the report is generated.
* **Token Management:** Real-time token counting ensures your input stays within the model's context window.
2. **💬 Medical Assistant Tab**
* **Text-Only Mode:** Chat with the MedGemma model about general medical concepts, differential diagnoses, or terminology without uploading an image.
---
### 🕹️ How to Use
#### For X-Ray Analysis:
1. Go to the **"🩻 X-Ray Analysis"** tab.
2. **Upload** a chest X-ray image (PNG or JPEG).
3. *(Optional)* Enter relevant **Clinical Information** in the text box.
4. Click **"🔬 Generate Report"**.
5. Once the report appears, use the chat box below it to ask **follow-up questions** (e.g., "Can you explain the pleural effusion finding?").
#### For General Questions:
1. Switch to the **"💬 Medical Assistant"** tab.
2. Type your medical question and hit Enter.
---
### 🧠 Model Information
- **Model:** `google/medgemma-4b-it`
- **Architecture:** MedGemma is built on top of Gemma 3, fine-tuned with medical instruction data.
- **Hardware:** Powered by Hugging Face **ZeroGPU** (dynamic H100/A100 allocation) for efficient inference.
- **Requirements:** This Space uses `sentencepiece` and `protobuf` for tokenizer handling.
---
### ⚠️ Disclaimer
This application is intended for **research and demonstration purposes only**.
* It should **not** be used for clinical decision-making.
* The model may hallucinate findings or miss critical anomalies.
* All generated reports must be reviewed and validated by a qualified medical professional.
---
### 🔧 Known Limitations
- **Image Formats:** DICOM files are not yet supported — please convert to PNG or JPEG before uploading.
- **Model Size:** This demo uses the **4B** parameter version. While fast, it may be less accurate than the larger 27B variant.
- **ZeroGPU Quotas:** Inference speed and availability depend on the current load of the ZeroGPU cluster.
---
Feel free to fork this Space to customize the system prompts or integrate it into your own clinical AI research workflows!