Related Experiment Video
Updated: Aug 9, 2025
07:11
Author Spotlight: Insights into Visual Cortex Research Through Wide-View fMRI Mapping
Published on: December 8, 2023
1.7K
Automatic uncertainty-based quality controlled T1 mapping and ECV analysis from native and post-contrast cardiac T1
Tewodros Weldebirhan Arega1, Stéphanie Bricq1, François Legrand1
1ImViA Laboratory, Université Bourgogne Franche-Comté, Dijon, France.
Medical Image Analysis
|February 24, 2023
Summary
A new quality control framework automatically detects inaccurate cardiac MRI segmentations using uncertainty. This ensures reliable T1 mapping and extracellular volume (ECV) analysis for diagnosing heart conditions.
Area of Science:
- Medical Imaging and Image Analysis
- Cardiovascular Disease Diagnosis
- Artificial Intelligence in Healthcare
Background:
- Deep learning for cardiac MRI segmentation yields state-of-the-art results but can produce errors.
- Inaccurate segmentations compromise downstream clinical tasks like myocardial tissue characterization.
- Automated quality control is crucial for reliable analysis of cardiac MRI data.
Purpose of the Study:
- To develop a fully automatic uncertainty-based quality control (QC) framework for cardiac T1 mapping and extracellular volume (ECV) analysis.
- To accurately detect failed cardiac MRI segmentations before clinical decision-making.
- To enable reliable characterization of myocardial tissues in healthy and pathological cases.
Main Methods:
- Cardiac structure segmentation using a Bayesian Swin transformer-based U-Net on native and post-contrast T1 mapping data (n=295).
- A novel uncertainty-based QC method employing image-level uncertainty features and a random forest classifier/regressor.
- Automatic computation of T1 mapping and ECV values post-QC for myocardial tissue characterization.
Main Results:
- The proposed QC method achieved a mean AUC of 0.927 for binary classification and a mean absolute error of 0.021 for Dice score regression.
- Outperformed state-of-the-art uncertainty-based QC methods, particularly in detecting segmentations from poor-performing models.
- Excellent agreement between automatic and manual T1/ECV values (Pearson coefficients 0.990 and 0.975) enabled accurate tissue characterization.
Conclusions:
- The developed uncertainty-based QC framework effectively identifies inaccurate cardiac MRI segmentations.
- This automated QC ensures reliable T1 mapping and ECV quantification for clinical decision support.
- Automatically computed T1 and ECV values can accurately characterize myocardial tissues across various cardiac diseases.
Keywords:
Bayesian deep learningCardiac MRI segmentationExtracellular volume (ECV)Native T1 mappingPost-contrast T1 mappingQuality controlVision transformer
