Harmonizing the Generation and Pre-publication Stewardship of FAIR bioimage data

Abstract

Together with the molecular knowledge of genes and proteins, biological images promise to significantly enhance the scientific understanding of complex cellular systems and to advance predictive and personalized therapeutic products for human health. For this potential to be realized, quality-assured bioimage data must be shared among labs at a global scale to be compared, pooled, and reanalyzed, thus unleashing untold potential beyond the original purpose for which the data was generated. There are two broad sets of requirements to enable bioimage data sharing in the life sciences. One set of requirements is articulated in the companion White Paper entitled “Enabling Global Image Data Sharing in the Life Sciences,” which is published in parallel and addresses the need to build the cyberinfrastructure for sharing bioimage data (arXiv:2401.13023 [q-bio.OT], https://doi.org/10.48550/arXiv.2401.13023). Here, we detail a broad set of requirements, which involves collecting, managing, presenting, and propagating contextual information essential to assess the quality, understand the content, interpret the scientific implications, and reuse bioimage data in the context of the experimental details. We start by providing an overview of the main lessons learned to date through international community activities, which have recently made generating community standard practices for imaging Quality Control (QC) and metadata (Faklaris et al., 2022; Hammer et al., 2021; Huisman et al., 2021; Microscopy Australia, 2016; Montero Llopis et al., 2021; Rigano et al., 2021; Sarkans et al., 2021). We then provide a clear set of recommendations for amplifying this work. The driving goal is to address remaining challenges and democratize access to common practices and tools for a spectrum of biomedical researchers, regardless of their expertise, access to resources, and geographical location.

Background and Motivation

Biological image data provides unique temporal and spatial data about biological molecules, which significantly enhances our understanding of complex biological systems. Moreover, sharing the vast amount of information captured in biological image data promises to advance human health and the economy, much the way sharing of genomics data and protein structure data has (i.e., the Human Genome Project, https://www.genome.gov/human-genome-project; and the Protein Data Bank - PDB, https://www.rcsb.org). Cellular, tissue, and medical imaging, for example, hold the answers to disease management (i.e., surveillance, prevention, diagnosis, and treatment), newly manufactured products, environmental resilience, and other global issues. Achieving this vision will require the interoperability, integration, and sharing of bioimage data across laboratories and research studies, thus maximizing the value of the billions of dollars invested annually in research around the globe.

Data sharing is globally recognized as highly desirable (UNESCO, 2022; UNESCO & Canadian Commission for UNESCO, 2022), but too often, publicly available imaging data lacks sufficient information to evaluate its validity and reproducibility independently and to assess its reusability (Botvinik-Nezer et al., 2020; Chen et al., 2023; Eriksson & Pukonen, 2018; Linkert et al., 2010; Marqués et al., 2020; Nature Editorial Staff, 2018; Pines, 2020; Sheen et al., 2019; Viana et al., 2023). Funding agencies have been working to improve the situation, for example, by requiring detailed data management and sharing plans (California Digital Library, 2022; European Research Council - Scientific Council, n.d.; NOT-OD-21–013: Final NIH Policy for Data Management and Sharing, n.d., Preparing Your Data Management Plan, n.d.). To improve reproducibility in research, many publishers now require more detailed materials and methods sections (Heddleston et al., 2021; Lee et al., 2024; Montero Llopis et al., 2021; Nature Editorial Staff, 2018; Schmied et al., 2023). In 2016, stakeholders from academia, industry, funding agencies, and publishing arrived at a set of so-called FAIR Data Principles—“Findable, Accessible, Interoperable and Reusable”—which are expected to expand the utility of data well beyond its original purpose (Wilkinson et al., 2016).

Several international community activities are engaged in addressing the challenges associated with the implementation of FAIR principles during bioimage data generation and post-acquisition processing (Supplemental Table 1). For many years, the Open Microscopy Environment (OME) has encouraged metadata collection and standardization of bioimage data file formats (Allan et al., 2012; Goldberg et al., 2005; Linkert et al., 2010; Moore et al., 2021, 2023; Swedlow et al., 2003, 2006). The global bioimaging community (see Supplemental Table 1) has more recently coalesced around improving standard practices for instrument QC and metadata (Faklaris et al., 2022; Hammer et al., 2021; Huisman et al., 2021; Microscopy Australia, 2016; Montero Llopis et al., 2021; Rigano et al., 2021; Sarkans et al., 2021) (Dekker et al., 2017, 2023); (Eriksson & Pukonen, 2018; Global BioImaging, 2015; Swedlow et al., 2021)(Ellenberg et al., 2018); (Strambio-De-Castillia et al., 2019);(Faklaris et al., 2022; Hammer et al., 2021; Huisman et al., 2021; Microscopy Australia, 2016; Montero Llopis et al., 2021; Rigano et al., 2021; Sarkans et al., 2021)(Abrams et al., 2020); (Eriksson & Pukonen, 2018; Global BioImaging, 2015; Swedlow et al., 2021); (Boehm et al., 2021; Nelson et al., 2021); (Faklaris et al., 2022; Hammer et al., 2021; Huisman et al., 2021; Microscopy Australia, 2016; Montero Llopis et al., 2021; Rigano et al., 2021; Sarkans et al., 2021)(Kemmer et al., 2023)(Faklaris et al., 2022; Hammer et al., 2021; Huisman et al., 2021; Microscopy Australia, 2016; Montero Llopis et al., 2021; Rigano et al., 2021; Sarkans et al., 2021). Through these community efforts, individual imaging laboratories and Shared Research Resources (or “core facilities”) are partnering with instrument manufacturers to define shared metadata frameworks and execute inter-laboratory studies to refine and deploy standard methods for QC (Abrams et al., 2023; Faklaris et al., 2022; Gaudreault et al., 2022; Nelson, 2022). These groups have demonstrated interest and a willingness to commit precious resources voluntarily, and the funding for these efforts has been sufficient to allow limited but significant headway within small pockets of the broader imaging community. Despite several remaining challenges, this progress is the beginning of a path forward for biomedical researchers to generate and manage reliable and well-documented microscopy data that can be trusted and reused. Satisfying FAIR data principles will then help unlock the vast potential of quantitative image-based research.

Most scientists agree with the FAIR data principles and intend to share and reuse bioimage data. Still, key technical hurdles remain, including the need for (i) cyberinfrastructure (Andreev et al., 2021)(Bajcsy et al., 2024);(Andreev et al., 2021)(Nagaraj et al., 2020; NIH Strategic Plan for Data Science, n.d.); (Andreev et al., 2021)(Nagaraj et al., 2020; NIH Strategic Plan for Data Science, n.d.), and (ii) standard metadata practices for the acquisition, analysis, management, and dissemination of bioimage data. Both are essential elements of good research data management (RDM) and stewardship (Boeckhout et al., 2018; Demchenko & Stoy, 2021; Steeleworthy, 2014).

Here, we summarize recent progress toward standard metadata practices for imaging data and recommend how this work can be extended to address remaining challenges and make best practices and software tools accessible to a wide spectrum of biomedical researchers regardless of their expertise, access to resources, and geographical location.

We start with challenges connected with data generation (i.e., sample preparation and image acquisition). We then describe issues related to the reliability and reproducibility of post-acquisition bioimage data processing, visualization, and quantitative analysis. Finally, we address the importance of data management and stewardship to ensure the link between image data and metadata is maintained across all aspects of the image-data lifecycle. Ensuring that the origin and lineage (i.e., provenance) of data can be tracked and its quality assessed is an essential prerequisite for guaranteeing FAIR data principles in microscopy. The scientific and sharing value derived from these metadata is the extent to which the associated bioimage data serves its intended scientific purpose and can be shared with other scientists to extract further insights.

Challenges and Recent Progress Associated with Data Generation

The metadata documentation of data generation lies at the very heart of the image data lifecycle, involves considerations made at the planning phase of the research project, is relevant before the sample hits the image acquisition platform, informs all subsequent image processing steps and is essential to ensuring the scientific value of shared image data.

Good practices in data generation and management that ensure that data are “FAIR from the start” are essential for rigorous and reproducible quantitative image-based biomedical research and for producing bioimage data that can be interpreted, trusted, and reused through model-based and data-driven mining, aggregation, reanalysis, and integrative modeling. It is crucial that third-party data users have ready access to all data-related information (i.e., metadata) that allows them to evaluate the suitability of given datasets for answering specific scientific questions before accessing or downloading them. This so-called “bioimage metadata” comprises two types of metadata: (i) data provenance metadata, which describes the experimental conditions, sample description and preparation, image acquisition (i.e., hardware description and image acquisition settings), and image processing, visualization, and analysis; and (ii) QC metadata, which describes system performance recorded through standardized QC protocols and metrics.

Challenges in Capturing Experimental Conditions and Sample Preparation Metadata

The description and interpretation of the results of any microscopy research project requires an extensive knowledge of the experimental steps preceding image acquisition and sample characteristics. This information should not only be captured in the Methods section of a paper (Larsen et al., 2023; Marqués et al., 2020; Montero Llopis et al., 2021) but it should also be made available as structured machine-readable metadata and include the following information: 1) the experimental procedures preceding image acquisition. 2) The origin of the biological sample and how it was obtained, experimentally treated, and prepared to produce the specimen for image acquisition. 3) The protocols and reagents (e.g., labeling procedures) used to visualize the structure of interest in the specimen. 4) the mounting technique and media used to preserve the integrity of the specimen during imaging. 5) the receptacle (e.g., slide and coverslip) used to hold the specimen during image acquisition.

Bioimaging communities are beginning to converge on essential metadata guidelines. For example, a 2019 community gathering to address data management and sharing in the light, electron, and X-ray microscopy fields resulted in a Recommended Metadata for Biological Images (REMBI) framework (Sarkans et al., 2021). REMBI provides a high-level map of metadata categories needed to ensure data interpretability and trust and can serve as a point of reference for different communities to converge on shared specifications. Minimum information guidelines for 3D microscopy, highly multiplexed tissue images and cell migration experiments have recently been developed to guide the documentation of experimental procedures and sample preparation (Cell Migration Standardisation Organisation, 2021; Hosseini et al., 2023; Reiff et al., 2022; Ropelewski et al., 2022; Schapiro et al., 2022). Compliance and implementation of metadata guidelines will depend on consistent ontologies for knowledge representation (Herr et al., 2023; Hotchkiss et al., 2019; Jupp, n.d.; Ong et al., 2017; Sickle Cell Disease Ontology Working Group, 2019), and where possible, the automated capture and annotation of metadata (FAIRdom, 2021; Hosseini et al., 2023; Musen et al., 2022; Wolstencroft et al., 2012).

Recent Advances in Capturing Microscopy Acquisition Provenance and QC Metadata

Quality assessment, reproducibility, interpretation, and reuse of image data require sufficient information about the hardware specifications, image acquisition settings, and performance of the instrument at the time of the data acquisition (Hammer et al., 2021; Huisman et al., 2021). A full technical description of the configuration of the imaging system can be used to calculate key information about spatiotemporal resolution, the noise associated with the system, and the physical and temporal dimensions of the image pixel data. An instrument performance assessment plan, including tracking standardized QC metrics at regular intervals, can be used to quantify changes in performance over time (Abrams et al., 2023; Faklaris et al., 2022; Gaudreault et al., 2022; Nelson, 2022). QC metrics allow us to quantify disparities between expected (theoretical) and observed (empirical) values and to compare with values measured at installation (i.e., t=0 of the microscope lifetime). Importantly, QC metrics also help to characterize and calibrate derived quantities extracted by image analysis (e.g., co-registration measurements). Ultimately, capturing the overall state (Rigano et al., 2021) and performance of a microscope as part of the metadata at the time of data acquisition is essential to identify potential batch effects in large datasets (Viana et al., 2023). Batch effects significantly affect the performance of Artificial Intelligence/Machine Learning (AI/ML) algorithms (Arevalo et al., 2023; Cimini et al., 2023; Tromans-Coia et al., 2023), so capturing instrument state and performance is critical for the interpretation of results (Chen et al., 2023; Viana et al., 2023).

Ensuring the Validity and Reproducibility of Image Data Visualization, and Analysis

A detailed workflow must be provided to independently assess the quality and reproducibility and ensure the interpretation of bioimage data analysis, as described in recently developed community guidelines (Aaron and Chew 2021; Miura and Nørrelykke 2021; Schmied et al. 2023). Additionally, the data size, computing hardware characteristics, and networking requirements should be part of the analysis metadata. This is not a trivial request. A data analysis pipeline typically involves a complex dependency chain of multiple software packages (for example, (Ahlers et al., 2023)) that has to be accurately described to ensure the independent assessment and reproducibility of results. As such, analysis metadata must constrain the version of each software component in the chain of dependency to avoid often significant changes. Accurately reporting all dependencies beyond the primary software in a manual fashion is unfeasible and is best performed using automated package managers (e.g., the pip freeze command).

Since image processing pipelines may rely on several tools, care must be taken to ensure that the intermediate and final results of processing and analysis pipelines and the associated metadata are stored in a harmonized and comparable manner across different software tools (Könnecke et al., 2015). To this end, developers have made important strides toward the use of containers (Bajcsy & Hotaling, 2020; González & Evans, 2019; Mitra-Behura et al., 2021; Schapiro et al., 2021), and workflow tools (Berthold, 2023; Di Tommaso & Floden, 2023; KNIME Community & bioml-konstanz, 2023; Stirling et al., 2021; Wollmann et al., 2017, 2023). Future work is needed to make these solutions more robust and to promote their universal adoption.

Suppose one could recreate an analysis environment using the same hardware, operating system, image analysis software, and all parameter settings. In that case, the analysis of the same dataset at two different locations should produce the same result. In practice, however, an image analysis software that uses randomness as part of its computations is unlikely to produce the exact same result; in this context, results within an acceptable margin of error (i.e., similar result) would be considered sufficient (PyTorch Consortium, 2023; Registration Overview — SimpleITK, 2024; TensorFlow Development Team, 2023). Finally, algorithms and implementations that utilize randomness require special care. This entails sharing additional information such as seed values and other software parameters. For example, when using deep learning, replicating results obtained by a retrained or new model requires access to the code and model weights, which should be shared using an interoperable file format across deep learning frameworks (e.g., the Open Neural Network Exchange, ONNX format, https://onnx.ai/).

Everyday Stewardship of Image Data and Metadata During Active Data Production

Data stewardship is an intrinsic and essential aspect of generating high-quality image data that is “FAIR from the start.” For this to happen, data stewardship must involve the entire lifecycle of the data, starting with the planning phase and continuing during experimental design and execution, sample preparation, data acquisition, post-processing, visualization, and analysis. In addition, data stewardship must outlive the research project to ensure that well-documented published data remains available for re-use, as detailed in the companion manuscript (Bajcsy et al., 2024). Specifically, correct data stewardship ensures that the conditions used to generate, process, analyze, and validate data are transparent, documented, propagated alongside the data and automatically reported to downstream users in both human-readable forms (i.e., scientific publications) (Heddleston et al., 2021; Larsen et al., 2023; Marqués et al., 2020; Montero Llopis et al., 2021) and structured machine-readable metadata frameworks (Moore & Strambio-De-Castillia, 2021; Soiland-Reyes et al., 2022; Solbrig et al., 2023). Cloud-ready data exchange formats (Moore et al., 2021, 2023; Swedlow et al., 2021) and standardized Application Programming Interfaces (APIs) that allow integration of images and results (Hammer et al., 2021; Moore, 2022; Moore et al., 2021, 2023; Rigano et al., 2021; Sarkans et al., 2021; Schapiro et al., 2022). This, in turn, ensures that data can be trusted, correctly interpreted, reproduced, and reused through data aggregation, mining, integrative modeling, and further analysis (including AI/ML). In addition, proper data stewardship is crucial to organize data, thus avoiding the waste of time and resources needed to re-generate data that has been lost or cannot be interpreted and, as a result, promote efficiency and sustainability (economic, environmental, and societal) (Budtz Pedersen & Hvidtfeldt, 2023; Meyn et al., 2022).

As such, effective data stewardship requires well-maintained, enterprise-grade, scalable, open-source, and democratized cyberinfrastructure (Andreev et al., 2021). This cyberinfrastructure should include the following; 1) Persistent Identifiers (PIDs) for research resources, individuals, publications and data (Brown et al., 2022a, 2022b; Cousijn et al., 2021; McCafferty et al., 2023); 2) shared file formats and associated APIs (Marconato et al., 2023; Moore et al., 2021, 2023); 3) the use of community-defined ontologies (Côté et al., 2010; Lomax, 2019)(Ciavotta et al., 2022; Khurana et al., 2023); 3) bioimage metadata specifications (Hammer et al., 2021; Sarkans et al., 2021; Schapiro et al., 2022); and 4) community defined Next Generation Metadata frameworks (Moore, 2022). This cyberinfrastructure should interface with Electronic Lab Notebooks (ELN) and Laboratory Information Management Systems (LIMS) and, whenever possible, automatically capture and propagate output metadata from all relevant instrumentation (including but not limited to robotic, microfluidics, and image acquisition hardware) (Marx, 2022a, 2022b). In summary, cyberinfrastructure should cover the following three interconnected aspects:

• WHAT information should be captured in Bioimage Metadata (i.e., develop community specifications for Experiment description, Sample preparation, Image acquisition, Image Processing, Visualization, and Analysis metadata; in particular, Image acquisition metadata should include hardware specifications, image acquisition settings, and QC protocols and metrics) (Hammer et al., 2021; Huisman et al., 2021; Sarkans et al., 2021; Schapiro et al., 2022).

• WHERE Bioimage Metadata should be stored (i.e., OME-NGFF and Next Generation Metadata with shared APIs) (Moore, 2022)(Moore et al., 2021, 2023)

• HOW Bioimage Metadata should be captured to facilitate metadata annotation, data curation, and seamless integration of all aspects of the imaging pipeline (i.e., integration with LIMS, ELNs, and hardware instrumentation; leverage community-specifications and Next Generation Metadata frameworks; ontology-enriched, REMBI-based, modular template spreadsheets; incorporate QC protocols and output metrics as image metadata) (Bukhari et al., 2018; Hammer et al., December, 9-12 2019; Kobayashi et al., 2019/Dec 10-11 2019; Kunis et al., 2021; NFDI4Plants Consortium, 2022; Rigano et al., 2021; Ryan et al., 2021; Sansone et al., 2012; Wolstencroft et al., 2012).

FAIR quality control in Recently Emerging Imaging Modalities

A set of recently developed imaging modalities are emerging as techniques of choice to quantify the spatial distribution of molecules and supramolecular structures at the subcellular, cellular, tissue, and organismal levels as reviewed in (Baysoy et al., 2023; Moffitt et al., 2022; Vandereyken et al., 2023)(Hickey et al., 2022; Kinkhabwala et al., 2022; Rivest et al., 2023). These techniques include spatial RNA profiling methods capable of resolving hundreds of probes at subcellular resolution using light microscopy, in situ sequencing (ISS), and capture assay. In addition, several highly multiplexed antibody-based imaging techniques have also been developed and commercialized. Cryogenic transmission electron microscopy (cryo-EM) provides images of biological matter in a frozen-hydrated, near-native state. The cryo-EM field is recognized as a good example of effectively annotating metadata and stewardship of raw and derived data (Sarkans et al., 2021). The cryo-EM community agrees that detailed metadata must be made publicly available alongside the data and that metadata standards must be reviewed regularly to ensure fitness and relevance to the evolving community needs (Chiu et al., 2021; Sarkans et al., 2021). Data sharing of cryo-EM derived (EMBL-EBI, n.d.-a; Lawson et al., 2011; RCSB Consortium, 2019)) and raw data (EMBL-EBI, n.d.-b; Ludin et al., 2016) - both with associated metadata - has already been established and is enforced by most publishers,

We advocate for a similar approach to be adopted for multiplexed RNA, protein, and multi-omic imaging datasets. The recent introduction of these methods, combined with their complexity and diversity, resulted in various workflows related to sample processing, reagent QC, image acquisition, image processing, and image analysis without agreed-upon community standards (Vierdag & Saka, 2024). The establishment and wide adoption of such standards are essential for cross-lab and cross-platform data interoperability and analysis, which is even more critical to the community since these data are expensive to acquire (Hickey et al., 2022; Quardokus et al., 2023; Vandereyken et al., 2023) and are associated with a number of consortia efforts (Jain et al., 2023; Rozenblatt-Rosen et al., 2020; Snyder & HuBMAP Consortium, 2019). Critical details related to sample preparation, reagent validation, and platform-specific imaging parameters must be recorded and shared using community-defined metadata. Here, we highlight specific challenges around harmonizing multiplexed image processing and reporting that limit current ability towards FAIR data (Wilkinson et al., 2016).

Recommendations Toward Global Image Data Generation and Stewardship Standards

All stakeholders can play a role in all aspects described. Manufacturers could provide automated access to the full community-defined technical descriptions of instruments and QC metrics as part of the metadata (Marx, 2022a, 2022b). Funding agencies could allocate funds to support the development of tools, protocols, and metadata standards by bioimaging communities to be implemented routinely by core facilities and users. Publishers should insist that all aspects of microscopy metadata (i.e., hardware specifications, image acquisition settings, and QC metrics) be a part of the data package. Investigators should adopt the best practices being established by the broader imaging research community, ensuring the data they generate has the necessary human and machine readable metadata to facilitate FAIR requirements for themselves and others. Lastly, if these practices are to be universally adopted, the development of resources in multiple languages need to be encouraged and supported.

In this section, we recommend steps to promote the production and stewardship of image data that is “FAIR from the start” and ready to be shared and reused. These recommendations are summarized in a to-do list for various stakeholders presented in Text Box 1.

Text Box 1 -. A to-do list for various stakeholders.

Towards Global Image Data Generation and Stewardship A to-do list for various stakeholders

More details on the action items below are provided in Supplemental Material.

1. Ensure the long-term sustainability of national and international bioimaging communities (e.g., ABIC, ABRF, BINA, CBI, GBI, LABI, and QUAREP-LiMi; see also Table 1 in supplemental materials), with the specific purpose of enabling recurring gatherings to coordinate (i.e., discuss, recommend, update) the development of:

   1. Consensus guidelines for quality control procedures and standards to encourage the implementation and reporting of QC protocols and performance benchmarks, including for imaging instrumentation.
   2. Shared metadata specifications, exchange frameworks, and tools to minimize barriers to adopting metadata guidelines for academic, government, and industry stakeholders.
   3. Shared computational cyberinfrastructure is needed to generate and manage image data before publication. This infrastructure should include well-documented and high-speed software tools, frameworks, computing and storage equipment, and networks. It will support all stages of the imaging process, from data annotation to image acquisition and analysis. Community-defined standards are essential to ensure transparency about the instruments and algorithms used.
2. Invest in core facilities (aka Shared Research Resources) and their Personnel from all backgrounds and regions to:

   1. Provide expertise in sample preparation, validation of staining protocols, image acquisition, and image analysis.
   2. Democratize access through shared resources and promote collaborations to facilitate access to advanced technology.
   3. Serve as pivotal hubs for disseminating expertise and user training on all topics essential for preparing FAIR image data for sharing and engendering maximum reuse value across resourced and under-resourced regions and communities.
   4. Develop strong connections with software development centers to ensure cyberinfrastructure’s usability, customization, and democratization for imaging pipeline automation.

3. Support the career and recognition of imaging scientists specializing in generating and stewardship of FAIR image data. These include core facility personnel, image data curators, bio-image analysis experts, and research software engineers.

4. In collaboration with vendors, develop and deploy automated methods to capture harmonized and consistent metadata documenting all steps of the imaging pipeline, from reagents used to generate image data to microscopy instruments and peripherals.

5. Promote the use of Persistent Identifiers (PID) for the FAIR description of research resources (e.g., reagents, instruments, core facilities) and outputs (e.g., datasets) to facilitate reproducibility and reuse and ensure that the personnel involved in the research enterprise are appropriately acknowledged.

6. Develop metrics that describe the qualities of resultant image data.

Data Generation Recommendations

Overcoming challenges related to the generation of image data that is “FAIR from the start” requires specific solutions that should be planned for and carried out by all interested stakeholders. To guide the development of these solutions, we provide the following specific recommendations:

• Promote the widespread adoption of persistent identifiers—institutions (e.g., Research Organization Registry- ROR) (Gould, 2023), core facilities (e.g., ROR and Research Resource Identifier - RRID) (Bandrowski, 2022), personnel (e.g., Open Researcher and Contributor ID - ORCID) (Haak et al., 2012; Shillum et al., 2021), reagents (e.g., RRID) (Bandrowski, 2022), microscope instruments (e.g., PIDINST) (Krahl et al., 2021; McCafferty et al., 2023; Stocker et al., 2020), and datasets (e.g., Digital Object Identifiers - DOI)—to enable the FAIR description of all scientific entities, to facilitate reporting and reproducibility and to recognize the essential roles of imaging scientists and core facilities in the research enterprise (Brown et al., 2022a, 2022b; Cousijn et al., 2021; McCafferty et al., 2023). This recommendation is backed by recent cost-benefit analyses (Brown et al., 2022a, 2022b) that revealed significant financial advantages associated with the adoption of PIDs which would be related to reductions in staff salaries, the time saved from tedious data entry, and in the facilitation of technology advances.

• Promote the collection of full technical descriptions of microscope hardware specifications, image acquisition settings, and QC protocols and metrics (aka Microscopy Metadata) that comply 4DN-BINA-OME (NBO-Q) microscopy metadata specifications (Hammer et al., 2021; Huisman et al., 2021) being developed by consensus by imaging scientists and instrument manufacturers (Marx, 2022a, 2022b). Technical descriptions captured in Microscopy Metadata must become obligatory aspects of the production of any image data; in their absence, image data cannot reliably be quantified, reproduced, or reused and ultimately loses scientific value even when it is shared. As such, Microscopy Metadata should be made transparently available to microscope users, automatically collected using community tools (Kunis et al., 2021; Kunis & Dohle, 2022; Rigano et al., 2021; Ryan et al., 2021), and encoded using shared metadata frameworks (Moore et al., 2021, 2023).

• Ensure that instrument maintenance and quality assessment are adequately supported to ensure that they become common practice at all core facilities and individual laboratories utilizing microscopes and regardless of local resource availability. Specific funding mechanisms should be considered to provide the necessary instrumentation, training and personnel or traveling metrology services for under-resourced areas. This will allow for the performance of instruments to be evaluated at regular intervals using community-defined metrology standards and QC procedures that are appropriate for each experimental question (Abrams et al., 2023; Gaudreault et al., 2022; Nelson, 2022). Additional metrics might need to be collected for specific types of experimental approaches and desired outcomes.

• Emphasize large infrastructure investments in core facilities and regional infrastructural hubs (Budtz Pedersen & Hvidtfeldt, 2023) who employ trained personnel, including imaging scientists, data stewards, image analysts, and research software engineers. Such shared infrastructure would increase efficiency and reduce costs by maintaining and assessing the performance of instruments; promote the dissemination of technological advances (hardware and software); facilitate user training; provide guidance for experimental procedures, image analysis, and data stewardship; and provide image analysis and RDM services to facilitate the deposition of FAIR data packages containing the appropriate image metadata to specialized bioimage repositories. These infrastructure investments should include providing legal support to review data, identify appropriate Creative Commons (CC) and Open-Source Software (OSS) licensing, and/or carry out Personal Information (PI) redaction (human subject data).

• Invest in the development of open-hardware devices (e.g., robotic devices, fluidics systems, environmental control devices, microscopes, etc.) to carry out all aspects of data generation as the most appropriate way to democratize advanced technology and to ensure the efficient use of resources.

Data Processing and Analysis Metadata Recommendations

Once acquired, images often require complex processing, visualization, and analysis steps to extract quantitative information about the signal intensity of a given label, as well as the location, morphological characteristics, association, and movement of biological entities. To ensure that the results of image processing and analysis pipelines are reproducible and ready for FAIR sharing, community-defined guidelines should be adopted (Bialy et al., 2021; Miura & Nørrelykke, 2021; Schmied et al., 2023). In particular:

• Image processing and analysis workflows should be shared in binary containers such as Docker, Singularity, or Podman, and include the complete software environment to ensure that all aspects of the pipeline remain identical for all users.

• The processing and analysis workflow should specify which aspects of the microscopy acquisition process (e.g., magnification, resolution, and signal intensity) can affect its execution and should, therefore, be encoded in metadata.

• All aspects of the processing and analysis pipeline—including but not limited to data structure and size, rendering and processing steps, algorithm version and input parameters, and computing and networking requirements—must be documented in both human- and machine-readable formats to ensure interpretability, reproducibility, and downstream reuse. Such metadata should be captured either in the image data file (e.g., information about image rendering and processing) or as part of the documentation of the workflow (e.g., algorithm version and input parameters).

• As such, it is imperative that analysis metadata be captured using community-defined metadata specifications and storage frameworks to ensure maximal efficiency with which this information can be extracted and tracked across all steps of the pipeline without the need for repeated time-consuming interactions with the image data itself.

Data Stewardship Recommendations

• The everyday stewardship of data and associated metadata throughout the entire lifecycle of quantitative imaging experiments is essential to ensure rigor, reproducibility, and the production of high-quality image data that can be interpreted and is ready to be reused according to FAIR principles.

• The generation and stewardship of FAIR image data require full transparency, management, and reporting of all information related to the conditions used for data generation (e.g., experimental conditions, sample preparation, and image acquisition), as well as processing and analysis (e.g., image analysis, and visualization).

• RDM cyberinfrastructure supporting the generation and pre-publication stewardship of high-quality FAIR image data should be made available to all biomedical researchers using microscopes as an essential prerequisite for image data sharing and reuse.

• Such imaging RDM cyberinfrastructure must include advanced computing and data repositories that provide integrated data stewardship, metadata annotation, visualization environments, processing pipelines, and analysis routines (including AI/ML). Different components must be connected via high-speed networks to expedite upload and download as needed.

• RDM cyberinfrastructure is best supported by easy-to-use, enterprise grade, robust, and continually maintained and supported open-source software to carry out all steps of the imaging pipeline from experimental procedure (i.e., LIMS or ELN) to image acquisition (i.e., Micro-Manager, Pycro-Manager, Python Microscope, etc.) to image processing, visualization, and analysis and data processing pipeline (i.e., CellProfiler, Fiji, napari, CellPose, etc.).

• When proprietary software and instrumentation must be used, community-defined standards must be used to ensure transparency regarding all relevant algorithms and input parameters as well as instrument configuration and performance.

• Collection and reporting of metadata must be based on community-defined standards, and it must occur at two highly interconnected levels:

  • Human readable, which is primarily related to Materials and Methods (Heddleston et al., 2021; Larsen et al., 2023; Marqués et al., 2020; Montero Llopis et al., 2021), and represent a subset of the information captured in Bioimage Metadata, enabling users to understand and describe the imaging experiment.
  • Machine-readable, which is captured in Bioimage Metadata (i.e., all information needed to understand the lineage - aka provenance - and quality of image data) and represents the complete technical description to ensure full quality, reproducibility, and reusability (Moore, 2022; Moore et al., 2021, 2023).
• To ensure Machine Readability:

  • Metadata should be encoded in community-specified frameworks to be associated with standardized image data file formats (Moore, 2022; Moore et al., 2021, 2023), or workflow documentation, and equipped with readily available software API to facilitate the transfer of information across the different steps of the imaging pipeline. For example, it should be possible to be put in an SQL database automatically. A potential framework that could fulfill these requirements is LinkML (Solbrig et al.,). It allows for easy authoring of metadata schemas in the YAML format, which can be exported into other formats. It is also able to create classes in various programming languages that can serve to validate metadata.
  • It is imperative to use specific annotation tools and automation at all aspects of the imaging pipeline to ensure that image processing, visualization, and analysis pipelines can leverage metadata. Particular emphasis should be given to (i) automated processes for microscope systems and peripheral components, including community-defined QC procedures to ensure optimal instrument performance (Hammer et al., 2021; Schapiro et al., 2022); (ii) automated metadata annotation at all phases of the image-data life cycle (Kunis et al., 2021; Rigano et al., 2021; Ryan et al., 2021); and (iii) integrated image processing, visualization, and analysis pipelines.
  • Metadata annotations should be backed by ontologies and knowledge graphs. Ontologies provide descriptions of the hierarchical relationship between concepts and can be used to make metadata machine-readable. Their role in the harmonization of knowledge and data in biomedicine is increasingly recognized, especially in the context of AI, where they can provide valuable constraints that make ML more efficient (Lomax, 2019). Tools such as the Ontology Look Up Service (OLS) (Côté et al., 2010) can be used to locate appropriate ontological terms, although further development is needed to help sort through duplicates and identify which terms should be used in different contexts.
• Quality, transparency, reproducibility, and reusability require standards defined by all imaging community stakeholders.

  • Recent advancements have made it clear that the community, when participating in bioimaging organizations, networks, and initiatives, is willing and ready to take on this challenge.
  • However, deliberate, directed, targeted funding is needed to ensure that ongoing standardization efforts can be expanded to cover all essential aspects of the imaging pipeline.

• In addition to being essential for generating standards, community organizations are ideal ways to promote the broader adoption of standards via education, training, and outreach. As such, novel funding models should be implemented to ensure the sustainability of these essential endeavors, promoting the advancement of quantitative imaging and science as a whole.

• Cyber infrastructure for RDM is expected to impact the economy beyond the biological research enterprise. Tested and trusted microscope QC protocols and commonly accepted and unambiguous reporting criteria will be beneficial to the many applications of quantitative imaging for cellular analysis, including pathology, cell therapies, and regenerative medicine.