mri
Multiplanar and volume review
Rotate the brain, move between linked slices and cut into the volume. Compare sequences at the same location.
Surface · Volume · MIP
Loading the interactive MRI volume…
OPEN SOURCE · BUILT FOR BRAIN MRI
A scan you can
work with.
Review linked 2D and 3D images, take measurements, record observations and define regions of interest. Deliver the viewer through your existing patient or clinician portal.
No viewer installation. MIT application code. Your choice of model.
REVIEW TOOLS
Image review and documentation
Close-ups from the actual application using an openly released MRI study. Explore the linked views, a measurement, a written note and a painted region.
Measurements and ROI statistics
Place a ruler or circular ROI. Inspect sequence statistics and export the measurements as CSV.
Physical geometry · CSV export
Location-linked clinical notes
Add your own text and a 2D or 3D location. Return to the observation and share it through a portable JSON export.
Browser drafts · JSON export
Manual segmentation
Paint and erase voxel masks on slices or with a 3D brush. Check them across views, undo a stroke and export your work.
Manual segmentation · Undo / redoPUBLIC RESEARCH DEMONSTRATION
Evaluate the viewer
Open a brain-masked T1 and FLAIR study from the OpenNeuro IDEAS II dataset. The same viewer handles your prepared institutional scans.
- Rotate the volume and explore the cuts.
- Place a ruler and compare linked slices.
- Write a note and paint a mask.
- Export your work and import it again.
IMAGING SERVICE INTEGRATION
From completed export to review
Use your existing completed-export route for an institutional pilot, then add the adapters your workflow needs.
Export
One completed DICOM study or verified NIfTI sequences.
Prepare
Check roles, align images and build brain-oriented maps.
Deliver
Provide the standalone viewer through your controlled portal.
Review
Use clinician notes and masks, with optional AI drafts.
The included export watcher is serial. PACS networking, SSO, centralized notes and scalable workers are integration work.
See the export contract →OPTIONAL MODEL INTEGRATION
Image access through MCP
Connect a compatible local or hosted model through an MCP client. The interface follows the study; inference runs on the hardware and runtime you choose.
MCP INTERFACE
Study tools and image evidence
Sixteen tools expose slices, region crops, mirrored comparisons, statistics, viewer focus and review annotations.
Your model↔MCP client↔MRI tools
Image review needs a vision-capable model and a client that handles image results and tool calls. A cloud provider receives the images sent to its model.
Configure the connection →MEDGEMMA & OTHER MODELS
MedGemma and compatible models
MedGemma is Google's medical image-and-text model family. You can also connect another compatible model through the same MCP tools.
4BMedGemma 1.5 model
16MCP tools
LocalOr approved hosted AI
Choose a suitable runtime: a GPU server, a local computer running a vision model through llama.cpp, or another supported deployment. NVIDIA hardware is an option; the viewer does not depend on a particular accelerator.
Model memory, image support and tool routing depend on the chosen runtime. The included bounded review runner is one example implementation; check its backend requirements separately.
Plan model integration →IMPLEMENTATION SCOPE
Requirements for an institutional pilot
Brain-focused analysis
The current asymmetry map compares aligned cerebral hemispheres. Other anatomy needs its own masks, alignment and evaluation.
Anatomical scope →Controlled image delivery
The viewer contains the scan. Your portal supplies access controls. Browser drafts remain local until exported.
Data flow and sharing →Application and model licensing
MIT code you can inspect, adapt and integrate. Model weights have their own terms. Infrastructure and integration still have costs.
Licensing →COMMON QUESTIONS
Practical questions
Does the patient have to install anything?
No dedicated viewer installation is needed to open a prepared HTML in a supported browser. The institution installs the preparation tools on its processing workstation.
Can I connect my own AI model?
Yes, through a compatible MCP client. A model that inspects MRI images needs vision support; the client must route tool requests and image responses. The model runtime and accelerator are separate from the browser viewer.
Does MedGemma run inside the browser?
Inference runs in a separate model runtime. The browser displays the scan, your annotations and saved model evidence. Use a local runtime or an institution-approved hosted service.
Can this replace our PACS or clinical reading system?
This release supports viewer delivery and a research pilot. PACS networking, authenticated records and clinical validation need additional integration.
Can every MRI use the asymmetry maps?
The implemented analysis is brain-specific. Other anatomy needs appropriate masking, alignment and independent evaluation.
Can I share a segmentation with another doctor?
Export notes and masks as JSON and import them into the same scan in another browser. Drafts do not automatically synchronize; DICOM SEG export and central records need additional development.
INTEGRATION DOCUMENTATION
Plan a review workflow
for your imaging service.
Public repository URL pending. Source & clone instructions · Contact the maintainer ↗