ezBIDS: Guided standardization of neuroimaging data interoperable with major data archives and platforms

The paper addresses the challenge of standardizing neuroimaging datasets to enable sharing, reproducibility, and replication. The Brain Imaging Data Structure (BIDS) provides a widely adopted framework for organizing multimodal brain imaging data, but converting raw data into BIDS typically requires knowledge of the BIDS specification, coding skills, and command-line tools. Existing converters often demand manual configuration (heuristics, pattern matching) or provide limited GUI support, creating barriers for researchers. The purpose of the study is to introduce ezBIDS, a web-based, no-installation tool that guides users through semi-automated conversion of imaging data and associated metadata (including task events) into BIDS, lowering technical barriers and improving adherence to standards while supporting interoperability with repositories and platforms like OpenNeuro.org and brainlife.io.

The authors situate ezBIDS within the context of BIDS and its extensions for multiple modalities (MRI, MEG, EEG, iEEG, PET, ASL, microscopy) and note the role of repositories and platforms (OpenNeuro.org, brainlife.io) in widespread adoption. Prior tools for BIDS conversion vary in required user expertise and level of automation, with many relying on command-line interfaces, coding, or configuration files (e.g., heudiconv, dcm2bids, bidsify, bidskit, Data2BIDS, BIDScoin). Some offer limited GUI support (e.g., Osirix/Horos plugins, web-based BIDS Toolbox, Flywheel heudiconv), but typically lack integrated guidance, events file conversion, QA checks, and seamless interoperability. The comparison highlights ezBIDS as distinctive in combining a web interface, no coding requirement, task events conversion, QA checks, embedded BIDS validation, interoperability with open repositories, and collaborative features via shareable session URLs.

ezBIDS is delivered as a Software-as-a-Service via a web interface hosted on secure cloud infrastructure. Users upload raw imaging data (DICOM or outputs of dcm2niix: NIfTI/JSON; compressed archives supported). The system decompresses files as needed, converts DICOM to NIfTI/JSON using dcm2niix, and invokes the ezBIDS Core to infer BIDS mappings. Core approach (Propose-and-Revise):

• Propose phase: Automated extraction of BIDS-relevant information from uploaded data via rule-based heuristics; outputs a comprehensive ezBIDS_core.json with inferred subject/session IDs, series groupings, data types, suffixes, entity labels, DICOM metadata, screenshots, and QA information.
• Revise phase: The web interface presents the proposed mapping for user review and edits through dropdowns and guided selections, enforcing BIDS rules. Key Core functions:

1. determine_subj_ses_IDs: Assigns subject/session IDs using unique DICOM fields (PatientID, PatientName, PatientBirthDate) when available; falls back to folder names or regex detection of sub-/ses- patterns; sessions inferred from AcquisitionDate/Time unless explicitly specified.
2. determine_unique_series: Groups images into unique series by matching SeriesDescription, ImageType, RepetitionTime, EchoTime, allowing tolerance for minor TE/TR precision differences and handling retro-reconstructions ("_RR" in SeriesDescription).
3. datatype_suffix_identification: Determines data type and suffix using three heuristics: (a) explicit tags in SeriesDescription (e.g., anat_T1w), (b) common protocol keyphrases (e.g., tfl3d for T1w), (c) other metadata (ImageType, EchoTime thresholds). If unresolved, sets datatype to exclude for user decision.
4. entity_labels_identification: Detects additional entities via regex in SeriesDescription and relevant metadata (e.g., EchoNumber for multi-echo to set echo- entities). Enforces required entities (e.g., dir for SE field maps). User workflow pages:

• Dataset Description: Prepopulates dataset_description.json fields (BIDSVersion, DatasetName); users add authors, acknowledgments, funding, license, and citation information.
• Subjects/Sessions: Review and adjust subject/session labels; reset options include PatientName, PatientID, or numerical zero-padded labels.
• Series Mapping: View series-level BIDS info (datatype, suffix, entities) and edit once per unique series to propagate across all instances. run labels are auto-determined later.
• Events conversion: Unique feature to convert arbitrary task events timing files (.csv, .tsv, .txt, .out, .xlsx; including E-Prime exports) to BIDS-compliant events.tsv. ezBIDS extracts column headers, lets users map them to BIDS fields (e.g., onset, duration), handles units (ms to s), and links events to corresponding BOLD runs by matching sub, ses, task, run from column names or file paths; placeholders provided for unresolved mappings for user correction.
• Dataset Review: File-by-file adjustments (e.g., exclude low-quality runs, add acq- labels). Visualization with screenshots and Niivue. Runs BIDS validation and ezBIDS-specific QA checks with errors (must fix for compliance) and warnings (recommendations, e.g., suspected incorrect dir label, short 4D runs suggested for exclusion).
• Pseudo-anonymization/defacing: Optional defacing for anatomical images using ROBEX+Quickshear (recommended) or pydeface, with pre-reorientation (FSL reorient2std) to standard orientation; notes pseudonymization limits.
• Participants info: Extracts and allows editing of phenotype fields (e.g., age, sex, handedness) for participants.tsv/json.
• Finalization and export: Assembles BIDS directory and filenames; runs bids-validator; allows download or direct upload to OpenNeuro.org or brainlife.io. Non-compliant outputs can be exported with expectation of later fixes. Session URL can be shared for collaborative review. Templates and metadata capture:
• ezBIDS_template.json: Downloadable after a session, captures user modifications and mappings; when uploaded with new sessions, auto-applies to Dataset Description, Subjects/Sessions, Series Mapping, Events, and Participants Info, expediting subsequent conversions.
• Guided metadata capture: Supports adding required/recommended metadata not extracted by dcm2niix (e.g., ASL fields such as BackgroundSuppression) with type checking and dependency handling (e.g., BolusCutOffTechnique required when BolusCutOffFlag present). Initially validated for MRI/ASL, with planned PET and MEG support. Security and deployment:
• Hosted on Jetstream2 Cloud (HIPAA-aligned). Anonymous sessions secured by JWT (RS256) revoked after 3 days; data encrypted via HTTPS (TLS 1.3/SHA256), stored up to 5 days then purged; services on private subnet; access restricted to authorized admins; activities logged.
• GDPR and data residency: For users outside the USA, recommended local pseudonymization/defacing before upload, or running a local ezBIDS instance via Docker from the GitHub repository (reduced functionality for direct repository uploads). Validation:
• ezBIDS validated on 30 shared datasets spanning multiple MRI data types and three scanner manufacturers (Siemens, Philips, GE); iterative fixes informed by errors and discrepancies.

• ezBIDS provides a no-install, no-coding, web-based workflow to convert neuroimaging data to BIDS, integrating semi-automated heuristics with guided user revisions.
• Unique support for converting heterogeneous task events timing files to BIDS events.tsv, including unit handling (ms to s) and automated linkage to BOLD runs via entity labels.
• Embedded BIDS validation and additional QA checks offer actionable errors and recommendations (e.g., required entities, suspected incorrect dir labels, short 4D runs), improving dataset quality.
• Interoperability: Finalized BIDS datasets can be downloaded or uploaded directly to OpenNeuro.org and brainlife.io; shareable session URLs enable collaborative curation.
• Template mechanism (ezBIDS_template.json) streamlines repeated sessions by reusing prior mappings and user customizations.
• Guided metadata capture extends beyond dcm2niix-extracted fields (e.g., ASL-specific fields) with type checking and dependency enforcement, enhancing compliance for newer BIDS extensions.
• Security: Runs on HIPAA-aligned cloud with time-limited tokens, encrypted transport, short data retention, and private networking.
• Validation across 30 datasets from three major MRI vendors and multiple data types demonstrated robustness of the mapping heuristics and workflow.
• Compared to other tools, ezBIDS uniquely combines a web interface, no coding requirement, events conversion, QA checks, validation, collaborative features, and interoperability; it also handles newer data types such as qMRI sequences.

The tool directly addresses barriers to adopting BIDS by eliminating installation and coding requirements and by providing guided, semi-automated mapping with built-in validation and QA. By supporting events.tsv generation and linking for task fMRI, ezBIDS enhances replicability and reuse of task-based datasets that often lack standardized timing files. Interoperability with open repositories and processing platforms promotes FAIR principles, enabling broader sharing and standardized analysis via BIDS-apps. The approach helps reduce variability in downstream processing and facilitates transparent, reproducible workflows. The discussion notes that while ezBIDS prioritizes a user-friendly web interface over a CLI, future enhancements may include a CLI to broaden applicability. Expansion to additional modalities (PET, MEG/EEG) and broader browser support are planned to further increase impact.

ezBIDS introduces a secure, web-based, guided BIDS conversion platform that requires no installation or programming, supports both imaging and task events data, embeds validation and QA, and interoperates with major repositories and platforms. The Propose-and-Revise framework and reusable templates streamline standardization across studies, lowering barriers to FAIR data practices and facilitating reproducible neuroimaging research. Future work includes extending support to additional modalities (e.g., PET, M/EEG), enhancing metadata coverage, offering a CLI option for advanced users, improving browser compatibility, and expanding deployment options and regional hosting to meet data governance requirements.

• Current development and validation emphasize MRI; broader support for other modalities (PET, M/EEG) is planned but not yet fully validated.
• Web interface has been primarily tested on Chrome and Firefox; other browsers (e.g., Safari, Edge) require further validation.
• Hosting currently on USA-based cloud infrastructure; users outside the USA must ensure compliance with local data governance (e.g., GDPR). Local defacing/pseudonymization or running a local instance may be necessary.
• Reliance on dcm2niix means some modality-specific metadata may not be auto-extracted; ezBIDS mitigates this via guided metadata capture but user input may still be required.
• Events file mapping depends on available identifiers in file paths/columns; user intervention may be needed for accurate linkage.
• Although bids-validator is embedded, ezBIDS can export non-compliant datasets at user discretion, requiring subsequent corrections.
• No command-line interface is currently provided, which may limit customization for some users.