MRIQC IQM Variability and Artifact Prediction in Multi-Echo fMRI
Poster No:
1071
Submission Type:
Abstract Submission
Authors:
Melanie Garcia1, Céline Provins2, Oscar Esteban3
Institutions:
1Mass. General Hospital, Harvard Medical School, Charlestown, MA, 2Lausanne University Hospital and University of Lausanne, Lausanne, Vaud, 3Lausanne University Hospital and University of Lausanne, Lausanne, VD
First Author:
Co-Author(s):
Introduction:
Assessing BOLD fMRI signal quality is essential for reliable statistical analyses. MRIQC (Esteban et al., 2017) is widely used for single-echo fMRI quality control, extracting image quality metrics (IQMs) per scan and generating visual reports to identify outliers at the group level.
For multi-echo (ME) fMRI, MRIQC produces IQMs for each echo, but their variation distributions across echoes and their relationship to artifacts in the final scan remain unclear. This study aims to address these gaps, identify aggregation strategies for IQMs, and predict artifact presence using machine learning.
ME-fMRI acquires data at multiple echo times, combining them to improve signal quality (Kundu et al., 2017). This technique has gained traction over the past decade for its ability to better disentangle artifact signals from BOLD signal of interest (DuPre et al., 2021).
For multi-echo (ME) fMRI, MRIQC produces IQMs for each echo, but their variation distributions across echoes and their relationship to artifacts in the final scan remain unclear. This study aims to address these gaps, identify aggregation strategies for IQMs, and predict artifact presence using machine learning.
ME-fMRI acquires data at multiple echo times, combining them to improve signal quality (Kundu et al., 2017). This technique has gained traction over the past decade for its ability to better disentangle artifact signals from BOLD signal of interest (DuPre et al., 2021).
Methods:
Data. We extracted IQMs from 124 BOLD fMRI runs with four echoes. The dataset (Provins et al., 2023) included three tasks (resting-state fMRI, breath-holding, and a positive control task), collected in all four phase encoding (PE) directions (AP, PA, RL, LR). Following our protocols (Provins et al., 2023; Hagen et al., 2024), author CP manually assessed all runs with MRIQC-generated visual reports (see artifacts in Figure 1).
Analysis. IQM values from MRIQC processing were compared between echoes using Pearson and Spearman correlations. Three methods (range, mean, median) were tested to combine echo-wise IQMs, considering task type and PE direction, to build binary classification models (SVM, Random Forest, MLP, Logistic Regression) for each artifact. The dataset was split into train (80%) and test (20%) sets. Grid search with iterative (10 repetitions) 2-fold cross-validation optimized model parameters and assessed performance with balanced accuracy. Dummy classifiers were also trained to get baselines of randomness.
Random feature permutation tests (10 iterations) were performed for each estimator to assess feature importance through mean decrease of balanced accuracy.
Analysis. IQM values from MRIQC processing were compared between echoes using Pearson and Spearman correlations. Three methods (range, mean, median) were tested to combine echo-wise IQMs, considering task type and PE direction, to build binary classification models (SVM, Random Forest, MLP, Logistic Regression) for each artifact. The dataset was split into train (80%) and test (20%) sets. Grid search with iterative (10 repetitions) 2-fold cross-validation optimized model parameters and assessed performance with balanced accuracy. Dummy classifiers were also trained to get baselines of randomness.
Random feature permutation tests (10 iterations) were performed for each estimator to assess feature importance through mean decrease of balanced accuracy.
Results:
Highly performant models and artifact-specific groups of important features were found.
67 models (MLP: 24; Logit: 20; RF: 14; SVM: 9) outperformed dummy models, with CV train/val. scores > 75%, train > 50%, and test = 100%. Wrap around was the most present target among these best models, and head motion seemed to be the best predicted artifact on average (see Figure 1).
Considering the top-3 models for each artifact, groups of features were highlighted as having predictive power. The important features depended on the type of artifact predicted (see Figure 1).
Linear combination of echo-wise IQMs and their summary statistics did not yield substantially better prediction of quality.
Most IQMs had correlations between echoes below 1, indicating non-linear relationships. While obtaining highly performing models (Figure 1), no combination scheme outperformed others in feature importance. These results suggest that more sophisticated combination approaches are necessary to better account for the non-linear nature of the IQM values between echoes while remaining generic to any number of echoes. Because ME data was also limited to four echoes here, we did not evaluate the influence of other ME setups, typically with three or five echoes.
PE direction has an influence on the range of IQM values between echoes, as well as on the predictive power of features for a same artifact.
For many IQMs, the range of values across echoes depends on PE direction (AP/PA or RL/LR) (see Figure 2). In addition, for each artifact, predictive features differed between models built on RL/LR scans and those on AP/PA scans, indicating variations in important feature groups based on PE direction.
67 models (MLP: 24; Logit: 20; RF: 14; SVM: 9) outperformed dummy models, with CV train/val. scores > 75%, train > 50%, and test = 100%. Wrap around was the most present target among these best models, and head motion seemed to be the best predicted artifact on average (see Figure 1).
Considering the top-3 models for each artifact, groups of features were highlighted as having predictive power. The important features depended on the type of artifact predicted (see Figure 1).
Linear combination of echo-wise IQMs and their summary statistics did not yield substantially better prediction of quality.
Most IQMs had correlations between echoes below 1, indicating non-linear relationships. While obtaining highly performing models (Figure 1), no combination scheme outperformed others in feature importance. These results suggest that more sophisticated combination approaches are necessary to better account for the non-linear nature of the IQM values between echoes while remaining generic to any number of echoes. Because ME data was also limited to four echoes here, we did not evaluate the influence of other ME setups, typically with three or five echoes.
PE direction has an influence on the range of IQM values between echoes, as well as on the predictive power of features for a same artifact.
For many IQMs, the range of values across echoes depends on PE direction (AP/PA or RL/LR) (see Figure 2). In addition, for each artifact, predictive features differed between models built on RL/LR scans and those on AP/PA scans, indicating variations in important feature groups based on PE direction.
·Figure 1
·Figure 2
Conclusions:
This study highlighted the need for exploring non-linear methods to combine IQMs across echoes and account for phase encoding direction, emphasizing the complex relationships that impact artifact prediction and quality assessment in ME-fMRI.
Modeling and Analysis Methods:
Activation (eg. BOLD task-fMRI) 1
Classification and Predictive Modeling 2
Exploratory Modeling and Artifact Removal
Keywords:
Computational Neuroscience
Data analysis
FUNCTIONAL MRI
Machine Learning
MRI
Other - Multi-Echo fMRI, Quality
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?
Provide references using APA citation style.
Esteban, O., Birman, D., Schaer, M., Koyejo, O. O., Poldrack, R. A., & Gorgolewski, K. J. (2017). MRIQC: Advancing the automatic prediction of image quality in MRI from unseen sites. PloS one, 12(9), e0184661.
Kundu, Prantik et al. (2017). ‘Multi-echo fMRI: A review of applications in fMRI denoising and analysis of BOLD signals’. In: NeuroImage 154, pp. 59–80. doi: 10.1016/j.neuroimage. 2017.03.033. url: http://dx.doi.org/10.1016/j.neuroimage.2017.03.033.
Provins, C., Lajous, H., Savary, E., Fornari, E., Franceschiello, B., Aleman-Gomez, Y., ... & Esteban, O. (2023). Reliability characterization of MRI measurements for analyses of brain networks on a single human. Springer Nature, 2023.