Mabrok Lab · AI for Healthcare
Where medical imaging AI is heading
An overview of papers from MICCAI 2024, 2025 and 2026, including the 2026 workshops and challenges. It shows which topics are growing and which are shrinking, which data and methods researchers rely on, and how much of the work shares its code.
Ask the field
Describe a research idea in plain English. We find the most closely related papers by meaning, not just keywords, and summarise what that corner of the field looks like: how active it is, which methods, data and groups drive it, and where to start reading.
Runs entirely in your browser: your question is never sent to a server. The first search downloads a small open AI model (about 34 MB, one time).
Topic briefing
Pick a topic for a one-screen summary: how fast it is growing, where it is applied, which datasets and groups drive it, and which papers to read first.
Trends: what is gaining and losing share
The share of main-conference papers that mention each topic in their title or abstract, comparing 2024 with 2026. A change in percentage points (pp) is shown, so small topics and large topics can be compared on the same scale.
Each topic, year by year
Topic map
Every paper from 2024–2026 and the 2026 workshops, placed so that papers with similar titles and abstracts sit close together, then grouped into 20 research clusters. Pick a cluster to highlight it and see how it has grown. Click a point to open the paper.
Collaboration network
Who works with whom at MICCAI 2026 (main conference and workshops). Each dot is a researcher with at least 3 papers, sized by paper count; lines join co-authors. Groups that publish together are detected automatically and described by the topics they focus on.
MICCAI 2026 landscape
The official subject areas authors chose for their papers, and which topics appear together in the same paper.
Data & open science
Which public datasets papers rely on, and how many papers share their code.
Peer review
Reviewers' initial scores for accepted main-conference papers, given before the authors' rebuttal, from the public reviews on papers.miccai.org (1 = strong reject, 6 = strong accept). Every paper here was accepted, so these scores show how confident reviewers were at first, not which papers were accepted.
MICCAI 2026 workshops & challenges
Workshops and challenges cover more specialised topics alongside the main conference. Here is how big each one is and which methods it uses most.
Paper explorer
Search and filter every paper. Each title links to its official open-access page, where you can read the abstract and reviews and download the PDF.
About this platform
Data
Paper titles, authors, subject areas, code and dataset links and reviewer scores come from the official MICCAI open-access site, papers.miccai.org, for MICCAI 2024, 2025, 2026 and the 2026 workshops and challenges. Workshop pages do not include code fields or reviews, so workshop papers are excluded from the peer-review statistics. This site does not host any papers. Every link goes to the official open-access page or PDF.
How topics are detected
A paper counts toward a method, task, modality or anatomy when its title or abstract matches a set of keyword rules (including well-known dataset names, e.g. "LIDC-IDRI" implies CT and lung). Negated mentions such as "without CT" are ignored. The same rules are used for every year, so trends are comparable.
We checked the rules by hand against a random sample of 60 MICCAI 2026 papers and 482 individual matches:
| Group | Precision | Recall |
|---|
Precision is the share of tagged papers that really are about the topic; recall is the share of papers about the topic that were found. When the two are close, missed and wrongly counted papers roughly cancel, so the shares are close to the true values. Treat them as approximate and focus on the direction of change.
How "Ask the field" works
Every paper's title and abstract was converted once into a numeric "meaning" vector with the open BGE-small embedding model. When you ask a question, the same model runs inside your browser (downloaded once from Hugging Face), turns your question into a vector, and ranks papers by similarity. Your question is never sent to any server, and the search costs nothing to run.
How to read the trends
"Share" is the percentage of that year's main-conference papers that mention the topic. "pp" means percentage points: a topic going from 5% to 12% of papers is +7 pp. A paper can count toward several topics, so the shares add up to more than 100%.
Built by
Mabrok Lab, AI for Healthcare. Last updated .