Exploring Brain MRI Variability and Harmonisation Approaches in the DPUK PET-MR Dataset
Poster No:
1533
Submission Type:
Abstract Submission
Authors:
Gaurav Bhalerao1, Pawel Markiewicz2, David Thomas3, Enrico De Vita4, Laura Parkes5, Gerry Thompson6, Jane MacKwen4, Georgios Krokos4, Catriona Wimberley6, Hallett William7, Li Su8, Stephen Smith1, Paresh Malhotra9, Nigel Hoggard8, John-Paul Taylor10, Craig Ritchie6, Joanna Wardlaw6, Paul Matthews9, Franklin Aigbirhio11, John O’brien11, Alexander Hammers4, Nick Fox12, Karl Herholz5, Frederik Barkhof3, Karla Miller1, Julian Matthews5, Ludovica Griffanti1
Institutions:
1University of Oxford, Oxford, Oxfordshire, United Kingdom, 2London South Bank University, University College London, London, United Kingdom, 3University College London, London, United Kingdom, 4King's College, London, United Kingdom, 5University of Manchester, Manchester, United Kingdom, 6University of Edinburgh, Edinburgh, United Kingdom, 7Perceptive, London, United Kingdom, 8University of Sheffield, Sheffield, United Kingdom, 9Imperial College, London, United Kingdom, 10University of Newcastle, Newcastle, United Kingdom, 11University of Cambridge, Cambridge, United Kingdom, 12University College, London, United Kingdom
First Author:
Co-Author(s):
Introduction:
The Dementias Platform UK's (DPUK) PET-MR Harmonisation Study acquired brain images of volunteers across 8 PET-MR scanners in the UK. With the goal of informing the design of future multi-centre clinical trials, improving image quality and consistency, it also represents an invaluable resource to evaluate new methods for imaging data harmonisation. PET and MRI scans were acquired from 37 healthy elderly participants divided into three experimental groups: (1) Repeatability – scans repeated on the same scanner (N=15), (2) Intra-manufacturer reproducibility – scans on the same scanner vendor (GE, Siemens) and model (GE SIGNA PET/MR, Siemens Biograph mMR) at different sites (N=10), and (3) Inter-scanner reliability – scans on different scanners vendors (GE and Siemens) (N=12). Our study aims to examine the variability of imaging derived phenotypes (IDPs) from T1-weighted (T1w) MRI scans and evaluate existing harmonisation approaches.
Methods:
All T1w scans were processed using the UK Biobank image processing pipeline (Alfaro-Almagro et al., 2018) to extract volumetric IDPs (total brain, tissue-specific, and hippocampal volumes). Variability was quantified as the Coefficient of Variation (CoV) for each subject pair and IDP.
Three harmonisation approaches were assessed on the inter-scanner group using CoV and variability ratio. Variability ratio was defined as the standard deviation within each subject (across scans) divided by the standard deviation of subject means across all subjects.
IQM-based (a): Image Quality Metrics (IQMs) from MRIQC (Esteban et al., 2017), CAT12 (Gaser et al., 2024) (See Bhalerao et al. 2024), and DSMRI tools (Kushol et al., 2023) were calculated. IQMs significantly differing across scanners (paired t-test, Bonferroni-corrected) with low correlation (-0.5<r< 0.5) to IDPs were identified. Principal components explaining 99% of variance were regressed from the IDPs.
IDP-based (b): Scanner-related batch effects were adjusted using the longitudinal ComBat method (Beer et al., 2020).
Image-based: Two techniques – c) Histogram matching, d) SynthSR were applied to harmonise T1w scans before running UKB pipeline (Iglesias et al., 2023).
Three harmonisation approaches were assessed on the inter-scanner group using CoV and variability ratio. Variability ratio was defined as the standard deviation within each subject (across scans) divided by the standard deviation of subject means across all subjects.
IQM-based (a): Image Quality Metrics (IQMs) from MRIQC (Esteban et al., 2017), CAT12 (Gaser et al., 2024) (See Bhalerao et al. 2024), and DSMRI tools (Kushol et al., 2023) were calculated. IQMs significantly differing across scanners (paired t-test, Bonferroni-corrected) with low correlation (-0.5<r< 0.5) to IDPs were identified. Principal components explaining 99% of variance were regressed from the IDPs.
IDP-based (b): Scanner-related batch effects were adjusted using the longitudinal ComBat method (Beer et al., 2020).
Image-based: Two techniques – c) Histogram matching, d) SynthSR were applied to harmonise T1w scans before running UKB pipeline (Iglesias et al., 2023).
Results:
Pre-harmonisation CoV was highest in the inter-scanner group (e.g., brain volume: 3.7±1.2%), moderate in the intra-scanner group (1.2±1.1%), and lowest in the repeated group (0.9±1.3%) (Fig. 1). Post-harmonisation (Fig. 2), CoV in the inter-scanner group decreased with IQM regression (e.g., brain volume: 3.7±1.2% to 1.5±0.9%) but showed high variability ratio for GM and hippocampus volumes, indicating reduced cross-subject variability and potential signal loss. Longitudinal ComBat further reduced CoV (e.g., brain volume: 3.7±1.2% to 1.4±0.8%) and overall performed best based on variability ratio. Image-based methods had mixed results: histogram matching lowered CoV for some IDPs (e.g., WM volume: 8.6±3.8% to 6.3±1.9%) but increased it for others (e.g., brain volume: 3.7±1.2% to 4.4±2.0%). SynthSR failed to effectively reduce CoV, primarily due to errors in downstream processes like brain extraction, as confirmed by visual inspection.
·Fig. 1. Comparison of image-derived phenotypes across repeated, intra-scanner, and inter-scanner groups.
·Fig. 2. Comparison of different harmonisation methods for volumetric IDPs in terms of A) CoV, B) Variability ratio
Conclusions:
The results highlight the promising performance of IQM- and IDP-based methods for multisite harmonisation. ComBat effectively mitigates within-scanner and cross-subject variability at the IDP level but depends on specifying a batch/site variable, limiting its applicability to repeated and intra-scanner groups due to dataset design (e.g., not all intra-scanner pairs share the same scanner/vendor). In contrast, IQM regression achieves comparable or greater reductions in within-scanner variability and offers greater versatility when scanner information is unavailable or highly heterogeneous. However, further optimisation is required to address cross-subject variability, guiding our future work on IQM-based methods. The inconsistent results from image-based approaches underscore the need for improved implementation and exploration of emerging deep-learning techniques.
Modeling and Analysis Methods:
Methods Development 1
Multivariate Approaches
Neuroinformatics and Data Sharing:
Workflows
Novel Imaging Acquisition Methods:
Anatomical MRI 2
Keywords:
Data analysis
Design and Analysis
MRI
Multivariate
NORMAL HUMAN
Statistical Methods
STRUCTURAL MRI
Workflows
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.
• Beer, J. C., Tustison, N. J., Cook, P. A., Davatzikos, C., Sheline, Y. I., Shinohara, R. T., & Linn, K. A. (2020). Longitudinal ComBat: A method for harmonizing longitudinal multi-scanner imaging data. NeuroImage, 220, 117129. https://doi.org/10.1016/j.neuroimage.2020.117129
• Bhalerao, G., Gillis, G., Dembele, M., Suri, S., Ebmeier, K., Klein, J., Hu, M., Mackay, C., & Griffanti, L. (2024). Automated quality control of T1-weighted brain MRI scans for clinical research: Methods comparison and design of a quality prediction classifier (p. 2024.04.12.24305603). medRxiv. https://doi.org/10.1101/2024.04.12.24305603
• 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. https://doi.org/10.1371/journal.pone.0184661
• Gaser, C., Dahnke, R., Thompson, P. M., Kurth, F., Luders, E., & the Alzheimer’s Disease Neuroimaging Initiative. (2024). CAT: A computational anatomy toolbox for the analysis of structural MRI data. GigaScience, 13, giae049. https://doi.org/10.1093/gigascience/giae049
• Iglesias, J. E., Billot, B., Balbastre, Y., Magdamo, C., Arnold, S. E., Das, S., Edlow, B. L., Alexander, D. C., Golland, P., & Fischl, B. (n.d.). SynthSR: A public AI tool to turn heterogeneous clinical brain scans into high-resolution T1-weighted images for 3D morphometry. Science Advances, 9(5), eadd3607. https://doi.org/10.1126/sciadv.add3607
• Kushol, R., Wilman, A. H., Kalra, S., & Yang, Y.-H. (2023). DSMRI: Domain Shift Analyzer for Multi-Center MRI Datasets. Diagnostics, 13(18), Article 18. https://doi.org/10.3390/diagnostics13182947