Related Experiment Video
Updated: Aug 24, 2025
08:19
Simultaneous Data Collection of fMRI and fNIRS Measurements Using a Whole-Head Optode Array and Short-Distance Channels
Published on: October 20, 2023
1.2K
Modeling human observer detection in undersampled magnetic resonance imaging (MRI).
Alexandra G O'Neill1, Emely L Valdez1, Sajan Goud Lingala2
1Mathematics Department, Manhattan College, Riverdale, NY, 10471, USA.
Summary
Assessing image quality in undersampled MRI is crucial. Model observers accurately tracked human performance in detecting brain metastases using FLAIR images, even with increased undersampling.
Area of Science:
- Medical Imaging
- Magnetic Resonance Imaging (MRI)
- Image Reconstruction
Background:
- Task-based assessment of image quality in undersampled MRI is vital for quantifying artifact impact.
- Fluid-attenuated inversion recovery (FLAIR) images are critical for detecting small brain metastases.
Purpose of the Study:
- To evaluate human observer performance in detecting small brain metastases using undersampled FLAIR images.
- To assess the impact of regularization parameters on image quality and task performance.
- To validate model observers (NPWE and S-DOG) against human performance in undersampled MRI.
Main Methods:
- Conducted two-alternative forced choice (2-AFC) studies with varying backgrounds for reconstructed FLAIR images from undersampled multi-coil data (4x undersampling).
- Employed a total variation (TV) constraint and analyzed its effect on human observer detection performance across a range of regularization parameters.
- Utilized non-prewhitening eye (NPWE) and sparse difference-of-Gaussians (S-DOG) with internal noise model observers to track human performance.
Main Results:
- Human observer detection performance remained stable for a broad range of TV regularization parameters, decreasing only at high values.
- The TV constraint did not significantly improve task performance in detecting small metastases.
- The S-DOG model observer with internal noise accurately tracked human performance across all studied regularization levels, while the NPWE model slightly overestimated performance at high regularization.
Conclusions:
- Model observers, particularly S-DOG with internal noise, can effectively track human detection performance in undersampled MRI tasks.
- This study demonstrates the first use of model observers to track human detection in undersampled MRI, offering a valuable tool for image quality assessment.
- Further research can refine model observers for more accurate prediction of human performance in complex imaging scenarios.
