ICASSP 2025 LF-MRI-challenge

Low-field MR Image Quality Challenge

 Sep. 2024 - Jun. 2025

 Hyderabad, India


The recent resurgence of portable, very low-field MRI systems to improve accessibility is constrained by image quality challenges compared to clinical scanners. The advent of machine learning methods provides a unique opportunity to address gaps in image resolution and signal-to-noise ratio (SNR). As very-low-field MRI gains the attention of the wider research community, this timely grand challenge invites participants from complementary backgrounds in signal processing, MRI analysis, machine learning, and related fields to break ground on this important topic of image quality. The challenge starts in Sep. 2024 and will last until Jun. 2025. To know more about the conference, visit the ICASSP website.

Overview

Mentor assisting attendees

Motivation

Two-thirds of the world does not have access to MRI. Low-field MRI improves accessibility but suffers from poor image quality

*Image generated using AI
Challenge_favicon

Tracks

Track 1 focuses on low-field image denoising while Track 2 advances low-field Super resolution reconstruction.

*Image generated using AI
Phantom low-field data

Data

3T paired data for phantom (47mT) and in vivo (0.3T)

*Image generated using AI

Challenge

General instructions - Participants can choose to pursue either or both challenge tracks and demonstrate it on the datasets described below. Each participant or team needs to host all source-code related to the challenge on a repository such as GitHub with an easy-to-use readme file containing instructions for installation and usage that enables reproducing their results. Participants are recommended to test their algorithms on both datasets. Top three teams for each track will be invited to present during the ICASSP conference.

Track 1: Image denoising - The denoising track will focus on developing, adapting, or integrating methods and algorithms that can increase the SNR by two-fold, i.e., reduce the noise by four times. While the use of deep learning algorithms has made significant progress in the denoising setting, the main challenge in the low-field setting is that algorithms lack multiple high-quality reference datasets that can be used to train these algorithms. Zero-shot, few-shot, transfer learning, and methods that do not require noiseless reference data may be better suited in this setting. The computation of the signal-to-noise ratio will follow the methods used in Ref. [3].

Track 2: Super-resolution reconstruction - Most MRI data used for segmentation and volumetry tasks are acquired at 1mm isotropic resolution. However, the low SNR at low fields often restricts the resolution to 1.5 x 1.5 x 5 mm3 to keep the scan time reasonable. While super-resolution reconstruction of natural images has advanced significantly, this lack of extensive high-quality low-field data makes this ill-posed problem a critical barrier to using portable MRIs for volumetric measurements, especially in the brain. The focus is to use ground truth data from phantoms and gold standard data from 3T acquisitions to train low-field super-resolution algorithms. The challenge is to use available very-low-field data comprising single or multiple orthogonal acquisitions to reconstruct a 1mm isotropic image volume with less than 5% error compared to the ground truth phantom data based on the vendor measurements.

Timeline

Sep. 7, 2024: Release of Dev Test set, Training dataset

Nov. 7, 2024: Release of Blind Test set

Nov. 23, 2024: Results of final subjective evaluation on Enhanced Blind Test set

Dec. 1, 2024: Grand Challenge 2-page Papers Due (by invitation only)

Dec. 16, 2024: Grand Challenge 2-page Paper Acceptance Notification

Dec. 23, 2024: Camera-ready Grand Challenge 2-page Papers Due

Jun.11, 2025: • Grand Challenge OJ-SP Papers Due (by invitation only)

Submissions

Please email your submission GitHub repo link to imr.framework2018@gmail.com

More submission details & leaderboard coming soon!

Organizers

Sairam Geethanath, Ph.D.

Mathews Jacob, Ph.D.

Hayit Greenspan, Ph.D.

FAQ

What is the format for final presentations?

To be announced shortly

Venue Details

ICASSP 2025 LF-MRI-challenge

Based on template by Niko Lazaris.