We welcome you to the world of vibe coding!

This is more than a tutorial. It is a space for the MICCAI community to explore how we actually code with LLMs.

As vibe coding becomes part of everyday research, many of us are experimenting with prompts, agents, scaffolds, and new debugging workflows. This tutorial provides an interactive forum to discuss what works, what fails, and how to make these workflows safer, more rigorous, and more reproducible.

Through practical sessions and open discussion, we will share techniques, experiences, tricks, and tips with one another.

Explore the stands

This tutorial is designed as a casual, interactive, multi-stand experience rather than a traditional lecture.

Participants can explore different stands, move between activities, and join the discussions that interest them most. Each stand presents a practical example of how modern LLM coding tools can support medical imaging research — from prototyping and debugging to testing, interpretation, and scientific communication.

Explore the interactive activities, example repositories, and supporting materials on our GitHub: https://github.com/VibeCodingMICCAI

01 — Review, Debug & Repair

Speakers: Yipeng HU, Shiqi HUANG 

Review an AI-generated annotation tool, identify hidden issues, design regression tests, and plan a controlled repair.

02 — Test, Readiness & Release

Speaker: Yaxi CHEN

Decide whether an AI-generated segmentation tool is ready for different research uses, based on the available evidence.

03 — From Research Idea to Proof of Concept

Speaker: Mattias HEINRICH

Use AI to explore prior work, identify datasets and software components, and break a new research idea into small, verifiable steps.

04 — Scientific Figure Generation

Speaker: Zuoou LI

Explore how AI tools can help generate, refine, and verify scientific figures for medical imaging research.

05 — Can You Trust the Explanation?

Speakers: Luisa Vargas, Matteo Pentassuglia, Alessandro Scavone, Eleonora Poeta & Maria A. Zuluaga

Investigate whether an AI-generated interpretation of experimental results is actually supported by the evidence.