Source-linked AI summary
Clinica: an open source software platform for reproducible clinical neuroscience studies
Alexandre Routier, Ninon Burgos, Mauricio Díaz, Michael Bacci, Simona Bottani, Omar El-Rifai, Sabrina Fontanella, Pietro Gori, Jérémy Guillon, Alexis Guyot, Ravi Hassanaly, Thomas Jacquemont, Pascal Lu, Arnaud Marcoux, Tristan Moreau, Jorge Samper-González, Marc Teichmann, Elina Thibeau--Sutre, Ghislain Vaillant, Junhao Wen, Adam Wild, Marie-Odile Habert, Stanley Durrleman, Olivier Colliot
TL;DR
Clinical neuroimaging studies are difficult to organize, process, and reproduce because they combine multimodal data, heterogeneous tools, and many analytical steps. Clinica addresses this gap with BIDS-based data handling, CAPS-organized outputs, automatic pipelines, and statistical and machine-learning tools, aiming to make such studies easier and more reproducible.
Problem
Clinical neuroimaging requires many processing and analysis steps across heterogeneous tools and datasets, making organization, sharing, and reproducibility difficult.
Method
Clinica combines BIDS input organization, public-dataset converters, CAPS outputs, automatic multimodal pipelines, and statistics, machine-learning, and deep-learning tools.
Results
Clinica provides a comprehensive platform intended to make clinical neuroimaging studies easier to perform and more reproducible.
Takeaways & Limitations
Standardized inputs and outputs, automated processing, and pipeline chaining support reproducible clinical neuroimaging workflows and connections to downstream analyses.
Takeaways & Limitations
Clinica lacks pipelines for several important functional neuroimaging modalities, including functional MRI and arterial spin labelling, and its processed-data quality control remains suboptimal.
Abstract
from arXiv · showhide
We present Clinica (www.clinica.run), an open-source software platform designed to make clinical neuroscience studies easier and more reproducible. Clinica aims for researchers to i) spend less time on data management and processing, ii) perform reproducible evaluations of their methods, and iii) easily share data and results within their institution and with external collaborators. The core of Clinica is a set of automatic pipelines for processing and analysis of multimodal neuroimaging data (currently, T1-weighted MRI, diffusion MRI and PET data), as well as tools for statistics, machine learning and deep learning. It relies on the brain imaging data structure (BIDS) for the organization of raw neuroimaging datasets and on established tools written by the community to build its pipelines. It also provides converters of public neuroimaging datasets to BIDS (currently ADNI, AIBL, OASIS and NIFD). Processed data include image-valued scalar fields (e.g. tissue probability maps), meshes, surface-based scalar fields (e.g. cortical thickness maps) or scalar outputs (e.g. regional averages). These data follow the ClinicA Processed Structure (CAPS) format which shares the same philosophy as BIDS. Consistent organization of raw and processed neuroimaging files facilitates the execution of single pipelines and of sequences of pipelines, as well as the integration of processed data into statistics or machine learning frameworks. The target audience of Clinica is neuroscientists or clinicians conducting clinical neuroscience studies involving multimodal imaging, and researchers developing advanced machine learning algorithms applied to neuroimaging data.
1 Inria, Aramis project-team, Paris, France
The passage identifies the Department of Neurology at the Institute for Memory and Alzheimer's Disease, Pitié-Salpêtrière Hospital, AP-HP, in Paris, France.
- The Department of Neurology is affiliated with the Institute for Memory and Alzheimer's Disease at Pitié-Salpêtrière Hospital, AP-HP.
- The listed location is Paris, France.
- The passage provides institutional affiliation information rather than study methods or findings.
12 Centre d’Acquisition et Traitement des Images (CATI), www.cati-neuroimaging.com
The supplied passages identify the ICM–Paris Brain Institute, its ARAMIS team at Pitié-Salpêtrière Hospital, and its Paris address, alongside keywords describing the paper’s topic.
- The affiliation is the ICM–Paris Brain Institute’s ARAMIS team at Pitié-Salpêtrière Hospital.
- The listed address is 47-83, boulevard de l’Hôpital, 75651 Paris Cedex 13, France.
- The keywords are neuroimaging, software, pipeline, data processing, machine learning, and multimodal data.
1 Introduction
Clinical neuroscience and machine-learning studies require many data-processing steps and expertise, while inconsistent data organization and heterogeneous software make workflows difficult to share, automate, and reproduce. Clinica addresses these challenges with automatic multimodal pipelines, dataset conversion, and tools spanning statistics and machine learning.
- Clinical neuroscience workflows require image preprocessing, measurement extraction, and statistical analysis across multimodal and often longitudinal data.
- Inconsistent dataset organization hinders data sharing, reuse, automatic pipelines, quality assurance, and integration across institutions.
- Nipype standardizes interaction with heterogeneous neuroimaging tools, while BIDS standardizes dataset organization and simplifies software execution.
- Multimodal studies remain challenging because researchers must master multiple modalities, tools, programming requirements, and methodological steps.
- Clinica provides automatic pipelines for T1w MRI, diffusion MRI, and PET, plus statistics, machine learning, deep learning, and public-dataset BIDS converters.
- Clinica targets clinicians, neuroscientists, and machine-learning researchers who need unified tools from raw data through analysis.
2 Clinica overview
Clinica is an open-source platform that combines established neuroimaging software into pipelines for multimodal clinical data and connects outputs to statistical and machine-learning analyses. It uses BIDS inputs, converts public datasets, documents pipelines, and organizes processed data for reproducible workflows.
- 2 Clinica overview: Clinica combines FreeSurfer, FSL, SPM, ANTs, MRtrix3, and PETPVC through Nipype for anatomical MRI, diffusion MRI, and PET processing.
- 2 Clinica overview: Clinica connects extracted features to statistical analysis with SPM and SurfStat and to machine learning with scikit-learn and PyTorch.
- 2 Clinica overview: BIDS is the expected input structure, and Clinica converts selected public neuroimaging datasets into BIDS format.
- 2 Clinica overview: The platform provides a schematic overview, a catalog of pipelines, usage examples, tool dependencies, and descriptions of generated outputs.
- 2 Clinica overview: Clinica’s MRI and PET pipelines combine software packages, accept BIDS data, convert public datasets, and store outputs in CAPS.
3 Clinica environment
Clinica uses BIDS for raw data and CAPS for processed outputs, with command-line tools organized around pipelines, conversion, data management, and development. This structure supports chaining analyses and sharing reproducible results.
- Software architecture: Clinica relies mainly on Nipype and distributes its Python package through PyPI for terminal-based installation and use.The package includes scientific Python dependencies such as NumPy, NiBabel, pandas, scikit-learn, nilearn, and PyTorch.
- Command-line interface: Clinica’s command-line interface includes commands for running pipelines, converting public datasets to BIDS, managing BIDS/CAPS data, and generating new pipeline templates.The main categories are clinica run, clinica convert, clinica iotools, and clinica generate.
- BIDS input: BIDS improves interoperability by using a unified file hierarchy that avoids database-management software and facilitates deployment across environments.Clinica expects BIDS-compliant inputs, with an interactive conversion option for cross-sectional datasets.
- Data organization: BIDS provides Clinica’s input structure, while CAPS organizes processed outputs separately and supports individual and group results.CAPS stores diverse outputs, including maps, meshes, deformation fields, and scalar measures, in a BIDS-like hierarchy.
- CAPS outputs: CAPS adds group-level organization so outputs from templates or population analyses can be identified with labels such as AD or HCvsAD.Individual outputs are stored under subjects, whereas group-study outputs are stored under groups with the participant set.
4 Image processing pipelines (clinica run)
Clinica provides automated pipelines for structural MRI, diffusion MRI, PET, machine learning, and deep-learning preparation. These pipelines transform multimodal data into standardized anatomical, diffusion, PET, and learning-ready outputs.
- Pipeline scope: Clinica’s image-processing pipelines cover T1w MRI, diffusion MRI, and PET, while its extensions support modular machine-learning classification and deep-learning model training.The pipeline collection includes modality-specific preprocessing, feature extraction, and learning-oriented preparation.
- Structural MRI: T1w pipelines perform affine registration, tissue segmentation, cortical-surface reconstruction, thickness estimation, parcellation, and longitudinal processing.The t1-linear, t1-volume, t1-freesurfer, and longitudinal FreeSurfer workflows produce voxel- and surface-based anatomical features.
- Diffusion MRI: Diffusion pipelines correct motion, eddy-current, susceptibility, and bias-field distortions before extracting tensor measures or connectomes.Clinica provides separate preprocessing workflows for acquisitions with phase-difference fieldmaps and for datasets without extra correction data.
- PET: PET pipelines normalize images and extract voxel-based or cortical-surface features, supporting amyloid and FDG tracers and optional partial-volume correction.The pet-surface workflows project PET signal onto cortical surfaces and can support longitudinal FreeSurfer processing.
- Machine and deep learning: Clinica supports modular SVM classification with spatial and anatomical regularization, and ClinicaDL provides training and validation tools for 3D medical-image deep learning.The spatial SVM approach regularizes using tissue maps such as GM, WM, and CSF.
5 Clinica utilities
Clinica utilities automate conversion of public neuroimaging datasets into BIDS and provide tools for checking and managing BIDS/CAPS datasets. Conversion requires downloaded source data but performs subsequent processing automatically.
- Dataset conversion: Clinica automatically converts ADNI, AIBL, NIFD, and OASIS datasets into BIDS using parallelized workflows without user intervention after download.The converters curate imaging and clinical data and can be extended to additional studies.
- Dataset conversion: The converters require study-specific imaging and clinical files, then produce a BIDS directory through a common command-line syntax.The supported converter names are adni-to-bids, aibl-to-bids, nifd-to-bids, and oasis-to-bids.
- Dataset management: Clinica provides BIDS/CAPS management commands for correcting centered NIfTI headers and checking missing modalities or processing outputs.These utilities support routine data-quality and organization checks.
6 Usage example
A Clinica example demonstrates a complete ADNI FDG-PET cortical-surface comparison between Alzheimer’s disease patients and healthy controls. The workflow converts data to BIDS, reconstructs surfaces, projects PET data, and runs covariate-adjusted statistics.
- Data conversion: ADNI T1w MRI and FDG PET data are converted into BIDS before processing with Clinica.The conversion command specifies the imaging and clinical directories and requests T1 and FDG PET modalities.
- Workflow: The example chains ADNI conversion, cortical-surface reconstruction, PET-surface projection, and statistical comparison into one reproducible workflow.Figure 3 summarizes the four-stage sequence from downloaded data to group comparison.
- Image processing: The t1-freesurfer pipeline extracts cortical surfaces, after which pet-surface projects FDG PET onto cortex and computes normalized PET measures.The PET command uses fdg as the acquisition label, pons as the SUVR reference region, and a PSF-information TSV for correction.
- Group comparison: The statistics-surface pipeline compares Alzheimer’s disease and healthy-control groups using age and sex as covariates.The resulting statistical outputs are stored in ADNI_CAPS/groups/group-ADvsHC.
7 Discussion
Clinica integrates established neuroimaging standards and tools into an open-source platform intended to make multimodal clinical neuroscience studies easier and more reproducible. Its scope is broad, but limitations remain in analytical flexibility, modality coverage, quality control, and deployment.
- Clinica addresses a gap between heterogeneous neuroimaging tools and users by offering unified pipelines for clinicians, neuroscientists, and machine-learning researchers.
- Clinica focuses mainly on T1-weighted MRI, diffusion MRI, and PET, while functional MRI and arterial spin labelling pipelines remain unavailable.
- Clinica combines BIDS inputs, standardized CAPS outputs, feature extraction procedures, and extensive software testing to support reproducible research.
- Users may face deployment barriers because Clinica depends on third-party software, requires command-line use, and currently lacks integrated advanced quality control.
- The post-hoc statistics and machine-learning tools cover commonly used approaches but do not include every possible analytical technique.
- Clinica provides comprehensive processing pipelines for different neuroimaging modalities and builds upon community standards and software tools.
9 Author Contributions
All authors contributed to study concepts and design, manuscript drafting or revision, and approval of the submitted final version.
- All authors contributed to study design, manuscript drafting or revision, final-version approval, and manuscript editing.
10 Funding
The research received support from French national initiatives and European Union H2020 projects.
- Funding came from several ANR and Initiative of Excellence programs, the PRAIRIE 3IA Institute, and European Union H2020 projects including EuroPOND and HBP SGA1.