Evaluating the performance of markerless prospective motion correction and selective reacquisition in a general clinical protocol for brain MRI

Abstract

Objective: Head motion is one of the most common sources of artefacts in brain MRI. When imaging young children, general anaesthesia is common, which is a limited resource. We evaluate the performance of markerless prospective motion correction (PMC) and selective reacquisition in a complete clinical protocol for brain MRI, comparing acquisitions with and without instructed intentional head motion.Materials and Methods: Image quality metrics and ratings were analysed for scans with and without PMC - acquired with six 2D- and 3D-encoded sequences in twenty-two healthy adults. The influence of PMC on motion-artefact-related changes in cortical thickness estimates was quantified using a general linear model.Results: For the motion-degraded 3D-encoded MPR and FLAIR sequences, image quality increased with PMC and reacquisition (p<0.001, corrected). Motionless scans with PMC showed slightly reduced (p < 0.05, corrected), but still diagnostic image quality. Cortical thickness estimates were widely correlated with motion level in the uncorrected scans (p<0.05), which was not apparent to the same extent in PMC scans. For the motion-degraded 2D-encoded TSE, STIR and DWI sequences we observed higher image quality with PMC and reacquisition (p<0.001, corrected), though the effect size varied. We did not observe an improvement in the T2* sequence (p>0.05, corrected), which is known to be sensitive to motion-related changes in B0. Discussion: Using PMC and selective reacquisition in sequences for standard clinical brain MRI improves diagnostic image quality when there is head motion.

Automated MRI Quality Assessment of Brain T1-weighted MRI in Clinical Data Warehouses: A Transfer Learning Approach Relying on Artefact Simulation

“…What's more, when it came to motion, we had to rely on previously used parameters (Loizillon et al, 2023(Loizillon et al, , 2024 since we could not find a reliable metric that did not require a reference image to quantify the amount of motion in a given MRI. Despite the recent efforts led by Eichhorn et al (2022) to find metrics that best correlate with radiological assessment, such as the average edge strength and Tenengrad measure, so far none of them have proven their robustness to other types of artefacts and in particular different types of contrast severity, leaving motion quantification as an open question.…”

“…For noise simulation, we explored various standard deviation (σ) ranges to find optimal parameters for corrupting MRIs with moderate and severe noise: σ = {[0, 10]; [5,15]; [10,20]; [15,25]; [20,30] Regarding motion artefacts, there is still no robust metric in the literature to quantify the motion present in MRIs. Recently, Eichhorn et al (2022) have shown that SSIM and PSNR were the metrics that correlate best with radiological assessment. What's more, they relied on the evaluation of pixel-by-pixel differences between images and were therefore very sensitive to misregistration between the original image and the corrupted image (Reguig et al, 2022).…”

“…These metrics also have the disadvantage of requiring a reference image, making it impossible to quantify motion artefacts within our manually annotated clinical routine images. Eichhorn et al (2022) also propose two different reference-free metrics for motion quantification (average edge strength and Tenengrad measure). However, due to their sensitivity to many cofactors such as contrast, neither of these metrics showed a significant correlation with our manual annotations (Figures A2, A3 and A4).…”

AI‐based motion artifact severity estimation in undersampled MRI allowing for selection of appropriate reconstruction models

“…The coil sentitivity maps were kept unchanged. We used 524 rigid‐body head motion patterns acquired on a scanner with an optical tracking system from a study where participants were instructed to perform shaking or nodding motion 22 . 393 patterns were used for training and 131 for validation.…”

“…We used 524 rigid-body head motion patterns acquired on a scanner with an optical tracking system from a study where participants were instructed to perform shaking or nodding motion. 22 393 patterns were used for training and 131 for validation. We multiplied each measured motion pattern by a uniformly randomly selected factor between 0 and 2 in order to generate a variety of training images with different amounts of motion.…”