AMDI Myocardial Infarction Dataset

Published: 5 May 2026| Version 2 | DOI: 10.17632/v2r59bz3tk.2
Contributors:
,
,
,
,

Description

This dataset provides late gadolinium enhancement cardiac magnetic resonance (LGE-CMR) images for the study of myocardial infarction (MI) analysis. The dataset was compiled to support the development and evaluation of image analysis methods including segmentation, intensity-based representation, and visualization of infarcted myocardial tissue. The data consist of 2D short-axis LGE-MRI images acquired retrospectively from 25 patients using a 3.0T MRI scanner. Two reconstruction types are included for each slice: magnetization-prepared inversion recovery (MR-IR) and conventional inversion recovery (CR-IR), providing variations in contrast distribution for the same anatomical location. All images are converted from DICOM to 16-bit TIFF format using DICOM rescale slope and intercept parameters to preserve the original intensity information. Three primary data components were provided. First, raw LGE-MRI images are provided without windowing or levelling, allowing users to apply custom preprocessing at their need. Second, a manually delineate and radiologist-verified multi-class segmentation ground truth masks are included with four RGB encoded class representing left ventricle myocardium, cavity, scar, and background. Third, intensity-encoded ground truth (IEGT) images are provided, where infarction regions are enhanced while preserving anatomical structure. IEGT were generated through a sequence of preprocessing steps including contrast normalization, unbiased non-local means denoising, contrast refinement via histogram matching, and mask guided intensity encoding. Additional supporting files included slice-level metadata in .csv format extracted from DICOM headers, and radiological reports in .pdf format describing patient-level findings. Dataset is organized at patient and slice levels with consistent naming conventions, enabling direct pairing between raw images, segmentation masks, and provided outputs. The dataset provides a structured resource for investigating both anatomical and intensity-based representation of myocardial infarction in LGE-CMR imaging.

Files

Steps to reproduce

Original source code and supplementary materials detailing the preprocessing steps in showing reproducibility are provided via GitHub repository link.

Institutions

Categories

Artificial Intelligence, Image Processing, Image Analysis (Medical Imaging), Applied Machine Learning

Funders

Licence