BIG-MRI @ MICCAI 2026 | Breast Imaging Group MRI Challenge

About

Can machines learn to predict cancer risk?

Breast MRI can reveal far more than what is visible at diagnosis. The BIG Challenge invites participants to develop AI models that predict 2- and 5-year breast cancer risk in women eligible for MRI screening, using MRI alone or combined with clinical information.

By advancing robust, clinically meaningful risk stratification, BIG-MRI challenge aims to support tailored screening decisions: identifying women who may benefit from closer follow-up while reducing unnecessary imaging for those at lower risk.

04Call for Papers

Submit Your Work

All participants are invited to submit a manuscript describing their methodology, solution, and challenge results. Manuscripts will be reviewed through OpenReview.

A detailed supplementary description of the full approach is mandatory for the top 5 teams on each leaderboard and should provide sufficient detail for reproducibility. For all other teams, supplementary material is optional. Supplementary material should be submitted in Word format and has no page limit.

Teams participating in both tracks only need to submit one manuscript through Track II.

Track I

Image-Based Prediction

Please download and use the manuscript template [here].

Length: 5–8 pages (excluding references)
Format: TBA
Review: Double-blind
Supplementary: Optional

Track II

Multimodal-Based Prediction

Please download and use the manuscript template [here].

Length: 5-8 pages (excluding references)
Format: TBA
Review: Double-blind
Supplementary: Optional

05Grand Challenge

The challenge is hosted on the Grand Challenge platform. Participants are invited to develop novel AI methods for predicting short- and longer-term breast cancer risk using breast MRI, with or without clinical information depending on the selected track. Participants will download the training data, develop and train their models locally, and submit only their final algorithms to Grand Challenge for evaluation. No model training will take place on the Grand Challenge platform.

Each track will have a separate leaderboard. Submitted algorithms will predict absolute 2-year and 5-year incident breast cancer risk per woman and will be evaluated on a held-out test set from the Netherlands Cancer Institute, together with an additional held-out set from Radboudumc.

Dataset

Dataset Introduction

We will release a dataset comprising of de-identified breast MRI exams from a very high-risk screening cohort at Radboudumc. The dataset will be distributed via Hugging Face in NIfTI format alongside a de-identified clinical table. More details about the data and relevant documentation will be released later. .....

....

Dataset & Baseline

Download: The dataset is hosted on Hugging Face: .NOT RELEASED!.
Baseline code: Necessary codes and templates will be shared via github.com.....

Competition

Two Challenge Tracks

The challenge is organized in two parallel tracks via the Grand Challenge platform. Participants may choose to focus on one or both. Both tracks will predict absolute 2- and 5-year incident breast cancer risk per woman, evaluated on the blinded held-out test set.

Challenge Track 1

Image-Based Risk Prediction

Given ....

GrandChallenge: Track 1 competition →

Challenge Track 2

Multimodal-Based Risk Prediction

Given...

GrandChallenge: Track 2 competition →

Track 1

Image-Based Risk Prediction

Participate: Track 1 on GC

Introduction

Model Input and Output

More information will be released in due time.

Evaluation Metrics

Please refer to the official challenge documentation for detailed evaluation specifications.

Track 2

Multimodal Risk Prediction

Participate: Track 2 on GC

Introduction

Model Input and Output

Evaluation Metrics

Please refer to the official challenge documentation for detailed evaluation specifications.

08Organizers

Organizing Committee

Lead Organizer

Breast Imaging Group MRI Challenge

Can machines learn to predict cancer risk?

Important Dates

Website Up

Dataset Documentation Release

Dataset Release

First Phase Submissions

Submit Your Work

Image-Based Prediction

Multimodal-Based Prediction

Submit via OpenReview

Dataset Introduction

Dataset & Baseline

Two Challenge Tracks

Image-Based Risk Prediction

Multimodal-Based Risk Prediction

Image-Based Risk Prediction

Introduction

Model Input and Output

Evaluation Metrics

Multimodal Risk Prediction

Introduction

Model Input and Output

Evaluation Metrics

Organizing Committee

Nika Rasoolzadeh

Tianyu Zhang

Xin Wang

Jonas Teuwen

Henkjan Huisman

Ritse Mann

Sponsors

Questions?