Adapting the MRIQC structural workflow for estimation of FLAIR scan quality
Poster No:
1852
Submission Type:
Abstract Submission
Authors:
Molly Ireland1, Heath Pardoe1, Perchyonok Yuliya2, Ricky Lu2, Greg Fitt2, David Vaughan1, Chris Tailby1, David Abbott1, Graeme Jackson1
Institutions:
1The Florey Institute of Neuroscience and Mental Health, Melbourne, Australia, 2Department of Radiology, Austin Health, Melbourne, Australia
First Author:
Co-Author(s):
Introduction:
MRIQC is a widely used software tool to generate automated assessments of MRI scan quality (Esteban et al., 2017a); however its structural workflow does not include FLAIR scans. FLAIR-specific challenges include low signal-to-noise (SNR) and less uniform voxel intensity values in the presence of white-matter (WM) hyperintensities and CSF suppression (Zwanenburg et al., 2010). This work adapted the MRIQC automated prediction tool for assessment of FLAIR scan quality.
Methods:
Whole brain 3D FLAIR MRI scans from the multi-site Australian Epilepsy Project (AEP, 5 sites, N= 235/25/7/3/2) and the OpenNeuro ds004332 dataset (N=29) were analysed (Ganz & Eichhorn, 2022). 30 scans were selected spanning a range of image quality, including scans with deliberate head motion, to develop and optimise the image processing pipeline (N = 20 AEP, 10 ds004332 scans). Our pipeline was validated on the remaining 271 FLAIR acquisitions. Scans were rated for overall scan quality from 1 (very poor) to 5 (excellent), by neuroradiologists and a trained rater.
We iteratively modified the MRIQC workflow and visually assessed tissue (CSF, WM, GM) and background segmentations. The workflow permutations targeted processing modules related to FLAIR scan properties e.g. SNR and contrast-to-noise (CNR). To evaluate our optimised FLAIR pipeline, we used linear regression to assess the relationship between visual ratings and 62 output image quality metrics (IQMs). The strength of these relationships was compared to the same regression analyses applied to T1w scans (AEP, N = 272, 5 sites) processed with the MRIQC (v23.1) pipeline.
We iteratively modified the MRIQC workflow and visually assessed tissue (CSF, WM, GM) and background segmentations. The workflow permutations targeted processing modules related to FLAIR scan properties e.g. SNR and contrast-to-noise (CNR). To evaluate our optimised FLAIR pipeline, we used linear regression to assess the relationship between visual ratings and 62 output image quality metrics (IQMs). The strength of these relationships was compared to the same regression analyses applied to T1w scans (AEP, N = 272, 5 sites) processed with the MRIQC (v23.1) pipeline.
Results:
Visual inspection of intermediate segmentations used to generate IQMs revealed that FLAIR-specific templates for spatial normalisation and brain tissue segmentation were important for good performance (Winkler et al., 2019). All regional segmentations showed significant improvements with the addition of symmetric normalization, implemented in the ANTs software package (Tustison et al., 2021). Brain tissue segmentation with ANTs Atropos improved using a nonparametric likelihood model (HistogramParzenWindows) to estimate voxel intensity distributions (Avants et al., 2011).
For artifact degraded scans, 3 processes significantly improved segmentation accuracy. Tissue segmentation improved by (1) modifying normalisation of WM voxel distributions and (2) mask refinement with morphological operations derived from the ANTs Brain Extraction script (Tustison et al., 2021). (3) Head mask segmentation improved with masking with a standard space head template converted to native coordinates. The final workflow modifications are shown in Figure 1.
Linear regression analyses showed the relationship between scan ratings and IQMs was comparable between the T1w and optimised FLAIR workflow (Figure 2), evidenced by the same direction of effects across the 13/15 IQMs, with statistically significant associations for both workflows. 44 IQMs from our FLAIR pipeline showed a significant association with scan ratings compared with 18 IQMs from the standard T1w MRIQC pipeline. For IQMs significantly associated with both scan types, the mean adjusted R² values were 0.20 and 0.04 for FLAIR and T1w pipelines respectively, suggesting greater explanatory strength for scan ratings in the FLAIR workflow.
For artifact degraded scans, 3 processes significantly improved segmentation accuracy. Tissue segmentation improved by (1) modifying normalisation of WM voxel distributions and (2) mask refinement with morphological operations derived from the ANTs Brain Extraction script (Tustison et al., 2021). (3) Head mask segmentation improved with masking with a standard space head template converted to native coordinates. The final workflow modifications are shown in Figure 1.
Linear regression analyses showed the relationship between scan ratings and IQMs was comparable between the T1w and optimised FLAIR workflow (Figure 2), evidenced by the same direction of effects across the 13/15 IQMs, with statistically significant associations for both workflows. 44 IQMs from our FLAIR pipeline showed a significant association with scan ratings compared with 18 IQMs from the standard T1w MRIQC pipeline. For IQMs significantly associated with both scan types, the mean adjusted R² values were 0.20 and 0.04 for FLAIR and T1w pipelines respectively, suggesting greater explanatory strength for scan ratings in the FLAIR workflow.
Conclusions:
We have developed a modified version of the MRIQC workflow for FLAIR scans and validated it across multiple sites and head-motion degraded scans. Our workflow shows a relationship between FLAIR IQMS and image quality ratings consistent with, and often exceeding, the MRIQC T1w workflow. As FLAIR acquisitions were standardized across AEP sites, further validation using more heterogenous datasets would be useful. Future research to identify FLAIR IQMs with high sensitivity to radiologist scan ratings could inform automated screening thresholds to detect low quality scans.
Modeling and Analysis Methods:
Exploratory Modeling and Artifact Removal
Image Registration and Computational Anatomy
Methods Development 2
Neuroinformatics and Data Sharing:
Workflows 1
Keywords:
Computing
Data Registration
Modeling
MRI
Open-Source Code
STRUCTURAL MRI
Workflows
Other - Quality Control; Quality Assurance; MRIQC
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
Esteban, O., et al. (2017b, June 25-29). MRIQC: Automatic prediction of quality and visual reporting of MRI scans [Poster Abstract]. 23rd Annual Meeting of the Organization for Human Brain Mapping (OHBM), Vancouver, BC, Canada. https://doi.org/10.7490/f1000research.1114419.1
Ganz, M., et al. (2022). Datasets with and without deliberate head movements for evaluating the performance of markerless prospective motion correction and selective reacquisition in a general clinical protocol for brain MRI [Dataset]. OpenNeuro. https://doi.org/10.18112/openneuro.ds004332.v1.0.0
Tustison, N. J., et al. (2021). The ANTsX ecosystem for quantitative biological and medical imaging. Scientific Reports, 11(1), 9068. https://doi.org/10.1038/s41598-021-87564-6
Winkler, A. M., et al. (2019). FLAIR templates [Data set]. Brainder. https://brainder.org/download/flair/
Zwanenburg, J. J. M., et al. (2010). Fluid attenuated inversion recovery (FLAIR) MRI at 7.0 Tesla: comparison with 1.5 and 3.0 Tesla. European Radiology, 20(4), 915–922. https://doi.org/10.1007/s00330-009-1620-2