DeepPrep: A deep learning-powered pipeline for 10-fold accelerated and robust MRI preprocessing
Poster No:
1587
Submission Type:
Abstract Submission
Authors:
Jianxun Ren1, Ning An1, Cong Lin1, Youjia Zhang1, Zhenyu Sun1, Wei Zhang2, Shiyi Li1, Ning Guo1, Weigang Cui1, Qingyu Hu3, Weiwei Wang4, Xuehai Wu5, Yinyan Wang6, Jiang Tao6, Tedd Satterthwaite7, Danhong Wang8, Hesheng Liu1
Institutions:
1Changping Laboratory, Beijing, Beijing, 2Peking University, Beijing, Beijing, 3Cornell University, Ithaca, NY, 4Beijing Normal University, Beijing, Beijing, 5Huashan Hospital, Fudan University, Shanghai, Shanghai, 6Beijing Tiantan Hospital, Capital Medical University, Beijing, Beijing, 7University of Pennsylvania, Philadelphia, PA, 8Massachusetts General Hospital, Harvard Medical School, Charlestown, MA
First Author:
Co-Author(s):
Introduction:
Neuroimaging is rapidly entering the era of big data with the launch of large-scale projects (e.g., UK Biobank (UKBB)), data-sharing initiatives (e.g., OpenNeuro), and international consortia (e.g., ENIGMA). Neuroimaging data typically require preprocessing using complex and multi-stage pipelines. However, the advancement of these pipelines lags behind the rapid expansion of data volume, causing significant computational challenges. Prevailing preprocessing pipelines, originally designed for smaller datasets, often require extensive processing times, ranging from hours to days for each scan (Ren, 2024a). At the same time, clinical applications demand fast turnaround and robustness at the individual level. To facilitate large-scale and translational neuroimaging studies, a computationally efficient, scalable, and robust preprocessing pipeline is urgently needed.
Methods:
We developed DeepPrep, a highly efficient, scalable, and robust preprocessing pipeline for structural and functional MRI (Ren, 2024a; Fig. 1). We incorporated state-of-the-art deep learning models to replace the most time-consuming steps, including FastCSR (Ren, 2022), SUGAR (Ren, 2024b), FastSurferCNN (Henschel, 2020), and SynthMorph (Hoffmann, 2024; Fig. 1a). To efficiently manage 83 discrete steps, we utilized the workflow manager Nextflow (Di Tommaso, 2017), to optimize computational resource utilization and flexible deployment across local computers, high-performance computing (HPC) clusters, and cloud computing platforms (Fig. 1b).
To comprehensively evaluate DeepPrep's performance, we applied it to 55,069 scans from 7 datasets with diverse populations, scanners, and imaging parameters, including the UKBB, Mindboggle-101 dataset, MSC dataset, CoRR-HNU dataset, and three clinical datasets. For comparison, we also preprocessed these datasets using fMRIPrep v24.0.0 under identical computing hardware environments (Esteban, 2019).
To comprehensively evaluate DeepPrep's performance, we applied it to 55,069 scans from 7 datasets with diverse populations, scanners, and imaging parameters, including the UKBB, Mindboggle-101 dataset, MSC dataset, CoRR-HNU dataset, and three clinical datasets. For comparison, we also preprocessed these datasets using fMRIPrep v24.0.0 under identical computing hardware environments (Esteban, 2019).
·Fig. 1 | DeepPrep is empowered by deep-learning models and the workflow manager.
Results:
DeepPrep achieved over 10-fold computational efficiency compared to fMRIPrep. For single-subject preprocessing of structural and functional MRI data from the UKBB dataset, DeepPrep required only 31.6 ± 2.4 min, significantly faster than fMRIPrep (Fig. 2a; 318.9 ± 43.2 min, p < 0.0001). In batch processing on a workstation, DeepPrep processed 1,146 subjects/week (Fig. 2b; average processing time = 8.8 min/sub), 10.4 times faster than fMRIPrep (110 subjects/week). On an HPC cluster, DeepPrep successfully processed the entire UKBB dataset (54,515 scans) in just 6.5 days, demonstrating its efficiency and scalability. On cloud platforms, DeepPrep's computational expenses were up to 22.1 times lower than those of fMRIPrep.
DeepPrep produced preprocessing results comparable to or superior to those of fMRIPrep in various metrics, such as anatomical parcellation, morphometric estimation, spatial normalization, temporal signal-to-noise ratio, task activation, and functional connectivity, confirming DeepPrep's accuracy while maximizing efficiency.
In clinical samples with distorted brains, DeepPrep showed significantly enhanced robustness. Among 53 clinical cases, DeepPrep demonstrated a 100.0% completion ratio and a 58.5% accuracy ratio, significantly outperforming fMRIPrep (Fig. 2c; completion ratio: 69.8%, p < 0.0001; accuracy ratio: 30.2%, p = 0.003). The failures and errors of fMRIPrep preprocessing were attributed to three key modules (Fig. 2d). Interestingly, these modules correspond to steps where deep-learning models replaced conventional algorithms in DeepPrep, highlighting the robustness of DeepPrep for clinical data processing.
DeepPrep produced preprocessing results comparable to or superior to those of fMRIPrep in various metrics, such as anatomical parcellation, morphometric estimation, spatial normalization, temporal signal-to-noise ratio, task activation, and functional connectivity, confirming DeepPrep's accuracy while maximizing efficiency.
In clinical samples with distorted brains, DeepPrep showed significantly enhanced robustness. Among 53 clinical cases, DeepPrep demonstrated a 100.0% completion ratio and a 58.5% accuracy ratio, significantly outperforming fMRIPrep (Fig. 2c; completion ratio: 69.8%, p < 0.0001; accuracy ratio: 30.2%, p = 0.003). The failures and errors of fMRIPrep preprocessing were attributed to three key modules (Fig. 2d). Interestingly, these modules correspond to steps where deep-learning models replaced conventional algorithms in DeepPrep, highlighting the robustness of DeepPrep for clinical data processing.
·Fig. 2 | DeepPrep achieves more than a 10-fold increase in speed and demonstrates robust performance in preprocessing clinical samples.
Conclusions:
We developed DeepPrep, an end-to-end, 10-fold accelerated, scalable, and robust pipeline for MRI preprocessing. The latest Docker image of DeepPrep is available on https://hub.docker.com/r/pbfslab/deepprep, with detailed documentation (https://deepprep.readthedocs.io). As fully open-source software, DeepPrep has the potential to become a cornerstone for high-throughput neuroimaging processing and clinical applications.
Modeling and Analysis Methods:
Methods Development
Motion Correction and Preprocessing 1
Neuroinformatics and Data Sharing:
Workflows 2
Informatics Other
Keywords:
Computing
Data analysis
FUNCTIONAL MRI
MRI
Open-Source Software
Spatial Normalization
STRUCTURAL MRI
Workflows
Other - deep learning
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. (2019). fMRIPrep: a robust preprocessing pipeline for functional MRI. Nature methods, 16(1), 111-116.
Henschel, L. (2020). Fastsurfer-a fast and accurate deep learning based neuroimaging pipeline. NeuroImage, 219, 117012.
Hoffmann, M. (2024). Anatomy-aware and acquisition-agnostic joint registration with synthmorph. Imaging Neuroscience.
Ren, J. (2022). Fast cortical surface reconstruction from MRI using deep learning. Brain informatics, 9(1), 6.
Ren, J. (2024a). DeepPrep: An accelerated, scalable, and robust pipeline for neuroimaging preprocessing empowered by deep learning. bioRxiv, 2024-03.
Ren, J. (2024b). SUGAR: Spherical ultrafast graph attention framework for cortical surface registration. Medical Image Analysis, 94, 103122.