BRAIN project: OpenNeuroPET: An Archive for PET data

National Institute Of Mental Health

Investigators

Linked publications, trials & patents

Abstract

In recent years, the importance of data sharing has increasingly been recognized by the neuroimaging community because of the poor replicability of findings, the need for appropriate quality control, the greater statistical power provided by larger samples, and the higher scientific impact of multilateral collaborations. Our funding bodies and scientific journals also increasingly encourage or require that the data be shared. However, data sharing can be clumsy and cumbersome in the absence of well-articulated guidelines, clear data formatting standards, and well-managed repositories. Building on the recommendations of leaders in the PET field, and in collaboration with other highly successful repositories of human neuroimaging data, Dr. Innis received NIH BRAIN initiative funding to establish a useful and freely-shared repository for human brain PET data (2021-2026). As a first step, Dr. Innis collaborated with Dr. Gitte Knudsen and a large international group of leaders in PET brain imaging to publish a paper (Knudsen et al 2020) that specified the content and format for publications and archives; these guidelines have been adopted worldwide, bringing much-needed cohesion to the field. Building on these guidelines, the goals of the BRAIN grant are threefold: 1) to establish the OpenNeuroPET Archive; 2) to support the use and adoption of the OpenNeuroPET Archive by reaching out to the community, maintaining a helpdesk, and ensuring compliance with international regulations; and 3) to establish and provide human molecular imaging brain atlases (MIBAs) that provide the mean three-dimensional distribution in brain of molecular imaging targets in aggregated samples of people. Four institutions are collaborating to accomplish these goals: the NIMH Intramural Research Program, Copenhagen University Hospital (in collaboration with Dr. Gitte Knudsen), Massachusetts General Hospital (in collaboration with Dr. Douglas Greve), and Stanford University (in collaboration with Dr. Russell Poldrack). We then completed a highly detailed specification of all the needed data. The document was approved by the BIDS Oversight Group as the accepted PET BIDS standard. Within the newly established OpenNeuroPET Archive, the data analysis pipelines for PET data require that all data be entered and stored in the proper format that can be seamlessly uploaded to the Archive. These data analysis pipelines are user friendly, validated by leaders in the field, and functional enough that researchers will choose to use them, regardless of whether they deposit data to the Archive. Our software systems are also built to circumvent errors by providing robust prebuilt tools for the most common steps in PET data analysis. The data analysis pipelines for the OpenNeuroPET Archive are comprehensive, aim to be applicable to all PET imaging centers, and are opensource. Salient features in active development include: 1) Conversion software and associated documentation to convert all primary data (PET, MRI, and plasma) into BIDS format, ensuring that data can be easily shared between centers. To better understand adoption patterns and guide future development, we have implemented telemetry and usage tracking in PET2BIDS. Notably, we have recorded PET2BIDS being used to convert 30,800 total scans between November 2024 and June 2025. We have also made ezBIDS deployable at both local (the user's computer) and site-wide levels (the latter runs at brainlife.nimh.nih.gov to address institutional data security concerns). 2) These pipelines standardize commonly taken preprocessing steps, such as motion correction, image co-registration, defacing, and time activity curve extraction. We have released software to handle these steps: petprep_hmc (head motion correction), petprep_extract_tacs (time activity curve extraction), and petdeface for co-registration and defacing. These pipelines have been tested on both published and pre-published data within the NIMH and have demonstrated a noticeable reduction in effort relating to the preprocessing and sharing of data. These preprocessing steps are being combined under a single pipeline within the larger ecosystem of NeuroImaging PREProcessing toolS (NiPreps), which will provide PET users with an equivalent to the popular fMRIPrep and sMRIPrep used by MR imagers. 3) Software for pharmacokinetic modeling of brain and plasma data using PETSurfer, along with additional modeling tools including Bloodstream and Dynamicpet. These tools provide researchers with robust, validated methods for analyzing PET data and extracting meaningful biological parameters. 4) Quality control datasets that can be used to confirm the reproducibility of the analytic pipeline. Quality control reports are also automatically generated by petprep_hmc, petprep_extract_tacs, and petdeface, providing researchers with comprehensive validation of their analysis workflows and ensuring data quality across different imaging centers. 5) MIBAs. Another key facet of the OpenNeuroPET Archive is the analysis and cross-subject averaging of the archived PET data to create atlases for each molecular target. The MIBAs can be displayed and used for quantitation and correlation with other imaging modalities, including MRI and fMRI. Our group has published two atlases and released NeuroMark, which provides an additional avenue to work with and create PET atlases. In anticipation of user concerns, these pipelines have been built as modular components so that users can integrate either individual parts or an entire pipeline into their analysis. When we receive feedback or additional test data from researchers, this modular approach enables us to more quickly iterate and release newer versions of our tools to accommodate users’ needs. These pipelines are packaged as self-contained applications that require only a BIDS dataset as input and can be run locally by any researcher with a minimal setup. To address these accessibility challenges, we have made significant progress in this reporting period by packaging all OpenNeuroPET tools as containerized applications using Docker and Singularity technologies. This approach enables flexible deployment options that address common institutional barriers. To make these pipelines more accessible, we sought to make them available without requiring download or installation. The PET BIDS conversion pipeline PET2BIDS was integrated into the web-based ezBIDS application (located at brainlife.io/ezBIDS) in 2023 to help address concerns commonly expressed by researchers regarding installation and usage of the command line based conversion pipeline. Additional OpenNeuroPET pipelines are currently being integrated into Brainlife.io to expand the range of PET applications available to researchers on that platform. We have enabled conversions to BIDS at the command line and provided all those not served by that solution a conversion option via ezBIDS; we continue to address user concerns with transmitting data "off site" and continue to work with Brainlife.io to support on-premise versions of ezBIDS and brainlife.io. We have also repackaged several of our pipelines to run "rootlessly", enabling compatibility with both the brainlife.io platform as well as high performance computing clusters. The NIMH infrastructure has been provisioned at brainlife.nimh.nih.gov as a testing/deployment environment to further support the interoperability of our applications with Brainlife.io. Finally, to further enable data sharing, we have released a new version of petdeface to help users fully anonymize their PET BIDS data. This tool addresses critical privacy concerns and enables researchers to share their data while maintaining participant confidentiality.