🚀 Try Me – One Click, Done Quick!

You can drop in a NIfTI file, watch every preprocessing step unfold in real time, explore the resulting connectome interactively, and ask questions in plain English or any natural language — all without leaving your web browser.

It pairs Streamlit’s reactive UI with LangChain’s tool-calling so you can see, tweak, and interrogate each stage of the pipeline:

skull-stripping → affine registration → tissue segmentation → AAL parcellation → graph construction → disease classification
fully interactive: 2D slice viewer, 3D Plotly volume, heat-map / graph visualizations, one-click downloads
pipeline orchestration by natural-language – e.g. run the pipeline without segmentation, enable network
RAG-powered Q & A over both your outputs and the UniBrain paper itself

⏳ We are working hard to enhance the tool, and a new version will be released soon.

A schematic overview of the UniBrain processing pipeline, illustrating the end-to-end workflow from input neuroimaging data (NIfTI/DICOM) through segmentation, registration, parcellation, network analysis, and final interactive chat-based querying.

🚀 Quick Start (Self-Deployment)

git clone https://github.com/soz223/UniBrainAssistant.git
cd UniBrainAssistant
python3 -m venv .venv && source .venv/bin/activate      # optional
pip install -r requirements.txt
export OPENAI_API_KEY="sk-..."                         # for LLM chat & RAG
streamlit run app.py

Open http://localhost:8501 → upload a NIfTI (or DICOM) → pick processing steps → Run. Then interact with your scan:

❯ skip segmentation             # run without the Segmentation step
❯ what does a high dice score mean?
❯ show only the network analysis stage

🔑 Environment Variables

Var	Purpose
`OPENAI_API_KEY`	Required for chat, command-parser LLM, and RAG
`IMG_SIZE`	(optional) override default 96³ voxel input size
`ATLAS`	(optional) specify brain atlas (e.g., “Desikan”, “AAL”)

📦 Core Dependencies

UI & Workflow: streamlit ≥1.32
Imaging & ML: torch, numpy, nibabel, SimpleITK, plotly, networkx
LLM & RAG: langchain, langchain-openai, faiss-cpu, openai (≥1.0)

See requirements.txt for exact version pins.

🤖 Command Grammar (for reference)

Intent	Examples (case-insensitive)
Skip step	`skip segmentation`, `no parcellation`, `without network`
Enable step	`enable analysis`, `turn on registration`, `enable classification`
Reset	`reset steps`, `reset pipeline`
General chat	anything else → routed to the UniBrain chat assistant

Processing flow:

Regex fast-path
if unresolved → GPT-4o-mini (CMD_SYS_PROMPT) → JSON command

🌐 Try the Demo Website

Visit the live demo at Ongoing Website
Enter your OPENAI_API_KEY in the sidebar API key field
Upload a .nii or .nii.gz file
Use the sidebar menu to enable/disable steps, then click Run

Chat with your data right in the browser:

❯ What can you do?
❯ Run the whole pipeline on the uploaded image
❯ Skip segmentation and do the whole preprocessing
❯ How does parcellation work?