[Paper Review] CMRxRecon: An open cardiac MRI dataset for the competition of accelerated image reconstruction

Chengyan Wang, Jun Lyu+26 more|arXiv (Cornell University)|Sep 19, 2023

Advanced MRI Techniques and Applications15 citations

TL;DR

This paper releases CMR×Recon, an open, large-scale cardiac MRI dataset with multi-contrast, multi-view, multi-slice and multi-coil data from 300 healthy subjects, plus processing scripts and baseline reconstructions to facilitate accelerated CMR image reconstruction research.

ABSTRACT

Cardiac magnetic resonance imaging (CMR) has emerged as a valuable diagnostic tool for cardiac diseases. However, a limitation of CMR is its slow imaging speed, which causes patient discomfort and introduces artifacts in the images. There has been growing interest in deep learning-based CMR imaging algorithms that can reconstruct high-quality images from highly under-sampled k-space data. However, the development of deep learning methods requires large training datasets, which have not been publicly available for CMR. To address this gap, we released a dataset that includes multi-contrast, multi-view, multi-slice and multi-coil CMR imaging data from 300 subjects. Imaging studies include cardiac cine and mapping sequences. Manual segmentations of the myocardium and chambers of all the subjects are also provided within the dataset. Scripts of state-of-the-art reconstruction algorithms were also provided as a point of reference. Our aim is to facilitate the advancement of state-of-the-art CMR image reconstruction by introducing standardized evaluation criteria and making the dataset freely accessible to the research community. Researchers can access the dataset at https://www.synapse.org/#!Synapse:syn51471091/wiki/.

Motivation & Objective

• Provide a publicly accessible, standardized raw k-space dataset for accelerated CMR image reconstruction.
• Include multi-contrast (cine and mapping), multi-view, multi-slice, and multi-coil data with manual segmentations.
• Facilitate fair comparison by offering preprocessing pipelines, evaluation criteria, and benchmark reconstructions.
• Offer baseline reconstruction scripts (GRAPPA, ESPIRiT, MoDL) and a platform for evaluation.

Proposed method

• Acquire 3T CMR data from 300 healthy volunteers with a 32-channel coil.
• Compress multi-coil data to 10 virtual coils for processing.
• Retrospectively undersample k-space (4x, 8x, 10x) in a Cartesian scheme and reconstruct with GRAPPA.
• Provide full k-space, undersampled k-space, masks, ACS lines, and reconstructed images in MATLAB and NIfTI formats.
• Manual myocardial and chamber segmentations are provided by an experienced radiologist.
• Supply scripts for state-of-the-art parallel imaging and model-based deep learning reconstructions as benchmarks.

Experimental results

Research questions

• RQ1 How can a standardized, publicly available CMR dataset accelerate development and benchmarking of accelerated reconstruction methods?
• RQ2 What are the baseline reconstruction performances (PSNR, SSIM, NMSE) for cine and mapping sequences under different undersampling factors?
• RQ3 How do multi-contrast and multi-coil data influence reconstruction quality and comparability across methods?
• RQ4 Can provided ground-truth segments support downstream analyses and evaluation of reconstruction-focused metrics?

Key findings

• The dataset includes 300 healthy volunteers with cine and T1/T2 mapping sequences, with 120 training, 60 validation, and 120 test cases for each modality.
• Undersampled k-space data (4x, 8x, 10x) and corresponding masks, ACS lines, and reconstructed images are provided for benchmarking.
• Benchmarks using GRAPPA, ESPIRiT, and modified MoDL are supplied, with quantitative metrics (PSNR, SSIM, NMSE) across multiple views and mappings.
• Manual segmentations of the myocardium and chambers are provided to support downstream evaluation and annotation.
• The data processing workflow converts scanner data to anonymized k-space in multi-coil format (emulated single-coil data provided) and offers ready-to-use processing scripts on GitHub and a Synapse evaluation platform.