Deepmriprep: Voxel-based Morphometry (VBM) Preprocessing via Deep Neural Networks
Poster No:
1502
Submission Type:
Abstract Submission
Authors:
Lukas Fisch1, Nils Winter2, Janik Goltermann3, Carlotta Barkhau2, Daniel Emden4, Jan Ernsting2, Maximilian Konowski5, Ramona Leenings2, Tiana Borgers6, Kira Flinkenflügel6, Dominik Grotegerd7, Anna Kraus3, Elisabeth J. Leehr7, Susanne Meinert7, Frederike Stein8, Teutenberg Lea8, Florian Thomas-Odenthal8, Paul Usemann8, Marco Hermesdorf9, Hamidreza Jamalabadi8, Andreas Jansen8, Igor Nenadić10, Benjamin Straube8, Tilo Kircher10, Klaus Berger9, Benjamin Risse11, Udo Dannlowski6, Tim Hahn2
Institutions:
1Institute for Translational Psychiatry, Münster, North Rhine Westphalia, 2University of Münster, Münster, Germany, 3University of Münster, Münster, North Rhine-Westphalia, 4University of Münster, Münster, North Rhine–Westphalia, 5University of Münster, Münster, Nordrhein-Westfalen, 6Institute for Translational Psychiatry, Münster, North Rhine-Westphalia, 7Institute for Translational Psychiatry, University of Münster, Münster, North Rhine-Westphalia, 8Department of Psychiatry and Psychotherapy, University of Marburg, Germany, Marburg, Hesse, 9Institute of Epidemiology and Social Medicine, University of Muenster, Muenster, Germany, 10Department of Psychiatry and Psychotherapy, University of Marburg, Marburg, Hesse, 11Faculty of Mathematics and Computer Science, Münster, North Rhine-Westphalia
First Author:
Co-Author(s):
Dominik Grotegerd
Institute for Translational Psychiatry, University of Münster
Münster, North Rhine-Westphalia
Institute for Translational Psychiatry, University of Münster
Münster, North Rhine-Westphalia
Elisabeth J. Leehr
Institute for Translational Psychiatry, University of Münster
Münster, North Rhine-Westphalia
Institute for Translational Psychiatry, University of Münster
Münster, North Rhine-Westphalia
Susanne Meinert
Institute for Translational Psychiatry, University of Münster
Münster, North Rhine-Westphalia
Institute for Translational Psychiatry, University of Münster
Münster, North Rhine-Westphalia
Frederike Stein
Department of Psychiatry and Psychotherapy, University of Marburg, Germany
Marburg, Hesse
Department of Psychiatry and Psychotherapy, University of Marburg, Germany
Marburg, Hesse
Teutenberg Lea
Department of Psychiatry and Psychotherapy, University of Marburg, Germany
Marburg, Hesse
Department of Psychiatry and Psychotherapy, University of Marburg, Germany
Marburg, Hesse
Florian Thomas-Odenthal
Department of Psychiatry and Psychotherapy, University of Marburg, Germany
Marburg, Hesse
Department of Psychiatry and Psychotherapy, University of Marburg, Germany
Marburg, Hesse
Paul Usemann
Department of Psychiatry and Psychotherapy, University of Marburg, Germany
Marburg, Hesse
Department of Psychiatry and Psychotherapy, University of Marburg, Germany
Marburg, Hesse
Marco Hermesdorf
Institute of Epidemiology and Social Medicine, University of Muenster
Muenster, Germany
Institute of Epidemiology and Social Medicine, University of Muenster
Muenster, Germany
Hamidreza Jamalabadi
Department of Psychiatry and Psychotherapy, University of Marburg, Germany
Marburg, Hesse
Department of Psychiatry and Psychotherapy, University of Marburg, Germany
Marburg, Hesse
Andreas Jansen
Department of Psychiatry and Psychotherapy, University of Marburg, Germany
Marburg, Hesse
Department of Psychiatry and Psychotherapy, University of Marburg, Germany
Marburg, Hesse
Benjamin Straube
Department of Psychiatry and Psychotherapy, University of Marburg, Germany
Marburg, Hesse
Department of Psychiatry and Psychotherapy, University of Marburg, Germany
Marburg, Hesse
Klaus Berger
Institute of Epidemiology and Social Medicine, University of Muenster
Muenster, Germany
Institute of Epidemiology and Social Medicine, University of Muenster
Muenster, Germany
Introduction:
Voxel-based Morphometry (VBM) has been used in over 7,000 studies to model associations between local brain tissue density and psychometric variables [1]. VBM toolboxes rely on compute-intensive preprocessing that segments each Magnetic Resonance (MR) image into a tissue map containing gray matter, white matter, and cerebrospinal fluid, and spatially normalizes this map to a template. This results in spatial alignment between the individual tissue maps, enabling the modeling of tissue density associations of individual voxels with generalized linear models (GLM). To address recent concerns about reproducibility [2], MRI datasets with thousands of individuals are used to increase the statistical power of VBM-based studies. Preprocessing these large datasets, which can contain up to 100,000 MR images [3], can take months with CAT12 [4], a leading preprocessing toolbox for VBM. Addressing this challenge, we evaluate a preprocessing pipeline, called deepmriprep, which performs all necessary processing steps with compute-efficient neural networks.
Methods:
For both tissue segmentation and spatial registration, neural networks are trained using 685 MR images compiled from 137 different OpenNeuro datasets [5], and their respective predictions are compared to silver ground truths. The Dice score is used to measure agreement with the ground truth tissue segmentation, while spatial registration is evaluated based on voxel-wise mean squared error. The regularity of the spatial registration is quantified via the linear elasticity. To ensure realistic performance measures, we employ a 5-fold cross-validation with images grouped by dataset, ensuring models are always validated with images from datasets unseen during training. All performance measures are compared to the existing CAT12 toolbox, and the lowest-performing predictions are visually inspected for potential weaknesses. Finally, the agreement of subsequent VBM-based statistical analyses is investigated by applying deepmriprep and CAT12 to 4,017 subjects, running GLM analyses, and correlating the resulting t-maps.
Results:
deepmriprep processes each MR image in just 4.6 seconds (see Figure 1), thereby achieving processing speeds 37 times faster than CAT12 while maintaining high agreement with the ground truth. Gray matter and white matter segmentation maps produced by deepmriprep show high Dice scores of 95.3 and 96.9, respectively. Even for MR images that yield incorrect or no tissue maps from CAT12, deepmriprep manages to produce reasonable results. During spatial registration, deepmriprep produces smooth, regular deformation fields comparable to those of CAT12. Although median performance metrics of the spatial registration slightly favor CAT12, visual inspection suggests the remaining performance gap is minor, even in challenging cases. GLM analyses based on deepmriprep and CAT12 preprocessing show high similarity, resulting in strong correlation coefficients of over 0.8 between their respective t-maps, even for psychometric variables with small effect sizes (see Figure 2).
·Processing time in seconds taken per MR image of deepmriprep and CAT12
·Absolute t-scores of GLM analysis between gray matter volume and six different psychometric variables based on deepmriprep- (left) and CAT12-preprocessing (right) thresholded at p < 0.001.
Conclusions:
The proposed neural network-based VBM preprocessing pipeline reduces the processing time required for the largest existing MRI study with more than 100,000 images from six months to just five days. Notably, this speed is achieved using a single consumer graphics card (NVIDIA RTX 3090). By alleviating the bottleneck of VBM preprocessing, deepmriprep enables laboratories without extensive computational resources to efficiently handle large-scale studies. deepmriprep can be conveniently installed as a Python package and is publicly accessible at https://github.com/wwu-mmll/deepmriprep.
Modeling and Analysis Methods:
Image Registration and Computational Anatomy 2
Methods Development 1
Segmentation and Parcellation
Keywords:
Computational Neuroscience
Machine Learning
Modeling
Morphometrics
Open-Source Code
Open-Source Software
Segmentation
Spatial Normalization
STRUCTURAL MRI
Other - Preprocessing
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.
Marek, S., Tervo-Clemmens, B., Calabro, F.J., Montez, D.F., Kay, B.P., Hatoum, A.S., Donohue, M.R., Foran, W., Miller, R.L., Hendrickson, T.J., Malone, S.M., Kandala, S., Feczko, E., Miranda-Dominguez, O., Graham, A.M., Earl, E.A., Perrone, A.J., Cordova, M., Doyle, O., Moore, L.A., Conan, G.M., Uriarte, J., Snider, K., Lynch, B.J., Wilgenbusch, J.C., Pengo, T., Tam, A., Chen, J., Newbold, D.J., Zheng, A., Seider, N.A., Van, A.N., Metoki, A., Chauvin, R.J., Laumann, T.O., Greene, D.J., Petersen, S.E., Garavan, H., Thompson, W.K., Nichols, T.E., Yeo, B.T.T., Barch, D.M., Luna, B., Fair, D.A., Dosenbach, N.U.F., 2022. Reproducible brain-wide association studies require thousands of individuals. Nature. 603(7902), 654–660.
Bycroft, C., Freeman, C., Petkova, D., Band, G., Elliott, L.T., Sharp, K., Motyer, A., Vukcevic, D., Delaneau, O., O’Connell, J., Cortes, A., Welsh, S., Young, A., Effingham, M., McVean, G., Leslie, S., Allen, N., Donnelly, P., Marchini, J., 2018. The UK Biobank resource with deep phenotyping and genomic data. Nature. 562(7726), 203–209.
Gaser, C., Dahnke, R., Thompson, P.M., Kurth, F., Luders, E., Alzheimer’s Disease Neuroimaging Initiative, 2024. CAT: a computational anatomy toolbox for the analysis of structural MRI data. GigaScience 13, giae049.
Markiewicz, C.J., Gorgolewski, K.J., Feingold, F., Blair, R., Halchenko, Y.O., Miller, E., Hardcastle, N., Wexler, J., Esteban, O., Goncavles, M., Jwa, A., Poldrack, R., 2021. The OpenNeuro resource for sharing of neuroscience data. eLife 10, e71774.