Lobe | Devpost

Inspiration

A brain MRI shows anatomy one slice at a time. For a student, understanding how those slices fit together means building a three-dimensional picture in their head. Studying with someone else adds another challenge: making sure both people are looking at the same place.

We built Lobe to make that exploration visual and shared. Our goal is simple: give learners a place to open a scan, explore its structure, and discuss what they see together.

What it does

Lobe is a collaborative workspace for exploring brain MRI scans in 3D.

Start with a prepared sample or upload four supported MRI sequences from a preprocessed study. Explore the scan through three slice views, then open the same case as an interactive 3D brain.

Rotate the model, adjust its transparency, separate its two halves, and reveal different tumor layers. A focus control brings the tumor region into view.

Invite a study partner to follow your exploration. They can follow your view, break away to inspect something independently, and rejoin when ready. Named pointers show exactly where someone is pointing, while shared chat keeps the discussion beside the scan.

Tumor outlines come from segmentation, which labels regions within an image. Uploaded studies use a pretrained model; prepared samples load previously generated results.

Lobe is an educational hackathon prototype. It is not clinically validated and does not provide diagnoses or treatment recommendations.

How we built it

We built the browser interface with React, Three.js, and NiiVue, connecting interactive 3D models with MRI slice views.

A Python imaging service validates uploaded scans, runs MONAI’s pretrained SegResNet model, and converts its output into 3D surfaces. Brain and tumor meshes use the same scan transform so they share a coordinate system. We used an existing model rather than training a new one.

Convex powers the collaboration. It stores sessions, annotations, and messages, while live queries keep participants updated. Shared state connects the presenter’s page, slice selection, camera, and display settings with participants who choose to follow.

Named pointers store a location on the model’s surface. Each participant maps that location into their own view, allowing the pointer to stay attached to the anatomy even when their display settings differ.

Challenges we ran into

Keeping spatial information consistent was a major challenge. MRI volumes, 3D surfaces, and shared pointers need explicit coordinate handling. A pointer also needs to move with the brain when its halves separate.

Collaboration required careful interaction design. Following a presenter should help learners without preventing independent exploration. We built an explicit follow-and-rejoin flow to support both.

We also had to distinguish useful interior inspection from unfinished features. Transparency and hemisphere separation work; a true cutaway view remains future work.

Accomplishments that we're proud of

We connected scan exploration and group discussion in one workflow.

Browser testing demonstrated two participants following the same case, exploring independently, returning to the presenter, seeing named pointers, and exchanging messages. That shared interaction is the center of Lobe.

What we learned

Detailed graphics alone do not make complex information understandable. Clear controls, a reliable way to recover your view, and a shared reference point matter just as much.

We also learned to distinguish visual alignment from medical correctness. Showing an outline in the right coordinate system does not prove its boundaries are accurate.

What's next for Lobe

Our priorities are reliable public demo access, complete scan-coordinate handling for annotations, and a true cutaway view.

We also want to verify Claude-generated session notes and calling across two devices, improve accessibility, and gather feedback from students and educators before expanding the feature set.