Should Researchers Always Preserve the Original Scientific Images?

If You Have Already Processed a Scientific Image, Do You Still Need the Original?

You have completed your scientific image analysis. The microscopy images have been denoised, the relevant structures segmented, and the final figures prepared for publication. The processed images are clearer, smaller, and easier to interpret than the original files.

Do you still need to preserve the raw images?

Perhaps the original files occupy considerable storage space. Perhaps the imaging software uses a proprietary format that is difficult to open. Or perhaps an AI enhancement tool has produced images that appear substantially better than the originals.

Deleting the source files may seem harmless once the analysis is complete. However, the original images may be the only surviving evidence of what the instrument actually recorded before processing, reconstruction, or interpretation changed their appearance.

The question is therefore not simply whether the original files might be useful again. It is whether the scientific record remains verifiable without them.

Original Scientific Images Should Generally Be Preserved as Essential Research Records

Why Original Scientific Images Matter and What Researchers Should Retain

What Counts as an Original Scientific Image?

An original scientific image is generally the earliest authentic image record produced or retained through the acquisition process, before subsequent analytical or presentation-oriented modifications.

Depending on the imaging modality, this may be a microscopy file, medical imaging dataset, instrument-generated image stack, camera raw file, or another native acquisition format.

However, the distinction between raw and processed data is not always straightforward.

Some instruments perform essential corrections, reconstruction, or calibration before producing an interpretable image. A medical scanner, for example, may generate reconstructed images from underlying measurement data.

Researchers should therefore identify which files constitute the primary acquisition records for their specific modality and which processing steps occur automatically within the instrument.

Where feasible and required, the retained record should include both relevant measurement data and the reconstructed images used in analysis.

The processed version may be scientifically useful, but it does not necessarily preserve every feature or measurement contained in the original record.

Why Is Preserving Original Images Important for Research Integrity?

Scientific images can function as primary evidence. Their credibility depends partly on whether the observations and subsequent transformations can be examined.

The U.S. Office of Research Integrity recommends preserving original image data and performing modifications on copies. Its image-processing guidance emphasizes that researchers should be able to establish what was recorded and how the image was subsequently altered.

Nature Portfolio's image integrity guidance similarly emphasizes accurate representation of original data and the retention of unprocessed image files.

Original images can help researchers determine whether a published feature was genuinely observed, introduced through processing, or inadvertently removed during enhancement.

They may also support the investigation of unexpected results, correction of analytical errors, and responses to legitimate questions raised during peer review.

Preservation therefore contributes to the traceability of research evidence rather than serving merely as an administrative obligation.

What Can Be Lost When Researchers Keep Only Processed Images?

Image processing can discard information that cannot subsequently be recovered from the final output.

For example, a contrast-adjusted image may contain clipped intensity values. A denoised image may no longer display weak signals. A compressed image may lose subtle variations that were present in the original acquisition.

Saving only the processed version can make it difficult to determine whether these changes affected the scientific interpretation.

The format alone does not determine whether an image is original. A TIFF file can be an authentic acquisition record or a heavily processed export. Provenance and processing history matter more than the filename extension.

Should Researchers Preserve Images That Were Not Included in the Final Paper?

Generally, relevant research images should be retained even when they do not appear in the published manuscript.

For example, a study may acquire hundreds of microscopy images but include only a few representative examples in its figures.

The unpublished images may still support the quantitative analysis, reveal variation across specimens, or provide evidence about how representative the published figures are.

Preserving relevant source images helps distinguish legitimate figure selection from selective presentation that conceals contradictory evidence.

However, the retention of every temporary file or technically unusable acquisition is not necessarily required in every research setting. Researchers should follow documented data-management and exclusion procedures, including institutional and funder requirements.

Images excluded from analysis because of quality-control failures may still warrant retention when they are needed to substantiate exclusion decisions.

Why Are Acquisition Metadata Important?

An image without relevant metadata may be difficult to interpret or reproduce.

Metadata can describe when and how an image was acquired, the instrument and settings used, and the spatial or intensity calibration needed for analysis.

For microscopy, relevant information may include objective magnification, pixel size, exposure time, illumination conditions, detector settings, and acquisition channels.

For medical imaging, relevant records may include modality-specific acquisition parameters, spatial geometry, and reconstruction information.

Some metadata are embedded in image files, while other information is stored separately in instrument software or laboratory records.

Researchers should preserve the information necessary to understand and evaluate the measurements, while handling identifying or sensitive metadata according to applicable privacy and security requirements.

What Is Image Provenance?

Image provenance describes the origin and processing history of an image.

It connects the final research figure or analytical result to the authentic acquisition record and the transformations applied between them.

For example, a published microscopy panel may have been derived from a native image through channel selection, background correction, cropping, and contrast adjustment.

A defensible provenance record should make it possible to identify the source image and understand how the displayed panel was produced.

A provenance record need not be unnecessarily elaborate, but it should be sufficiently complete to reconstruct consequential processing decisions where feasible.

Why Is Preservation Especially Important When AI Is Used?

AI image processing can introduce transformations whose effects are difficult to evaluate from the final image alone.

For example, AI denoising may suppress weak genuine signals, while AI upscaling may introduce model-dependent details.

If the original image is unavailable, researchers may be unable to establish whether a particular structure was present before enhancement.

Similarly, generative inpainting can replace missing regions with plausible visual content that was never observed.

Retaining the original acquisition record is therefore essential for distinguishing measured information from algorithmic estimates.

Researchers should also preserve relevant model and processing information, since the same image may produce different outputs under different software versions or settings.

Should Researchers Preserve AI-Generated Segmentation Masks?

Yes, when the masks form part of the analytical record needed to substantiate the results.

Segmentation masks may determine cell counts, object areas, tissue volumes, or other measurements reported in a study.

Preserving only the final numerical values may make it difficult to investigate whether incorrect boundaries or missed objects influenced those measurements.

Researchers should consider retaining the original images, model-generated masks, corrected masks, and relevant measurement outputs.

The need is particularly important when AI segmentation outputs are treated as research measurements.

Where storage limitations prevent retention of every intermediate file, researchers should follow an approved data-management strategy that prioritizes the records needed to reproduce or verify the analytical results.

Should Researchers Preserve Every Intermediate Image Version?

Not necessarily. The appropriate retention strategy depends on the analytical workflow and applicable requirements.

For deterministic processing, it may be possible to reproduce certain intermediate outputs from the original images, software, scripts, and recorded parameters.

However, some AI-based transformations may be difficult to reproduce exactly because of model availability, nondeterministic computation, or changes in software versions.

In such cases, retaining consequential intermediate outputs may be important.

Researchers should prioritize original data, final analytical inputs, scientifically significant transformations, and records needed to reconstruct the reported results.

The objective is a verifiable research record, not indiscriminate accumulation of every temporary file generated by software.

Can Researchers Delete Original Images After Publishing a Paper?

Publication does not ordinarily eliminate applicable research-data retention obligations.

Institutions, funders, journals, and national regulations may require research records to be retained for specified periods after project completion, publication, or another defined event.

For example, the U.S. Public Health Service research misconduct regulation at 42 CFR Part 93 establishes certain record-retention requirements in the context of research misconduct proceedings. Those requirements should not be confused with a universal retention period for all research images.

Other research sponsors and institutions maintain separate requirements.

Researchers should consult the applicable policies before disposing of original data. Additional obligations may arise when an investigation, audit, litigation hold, or unresolved research concern is pending.

How Long Should Original Scientific Images Be Retained?

There is no single retention period that applies universally across all scientific disciplines and jurisdictions.

The required period may depend on institutional policy, funder agreements, regulatory requirements, participant consent, intellectual property considerations, and the nature of the research.

Some institutions use minimum retention periods measured in years, while regulated clinical or other specialized research may require different arrangements.

Researchers should not assume that a commonly cited period, such as five or seven years, automatically applies to their project.

Retention periods may also begin from different events, including the completion of research, publication, or termination of a funding agreement.

The appropriate approach is to establish the applicable retention schedule before data collection and review it before any planned disposal.

Can Storage Limitations Justify Deleting Original Images?

Storage limitations are a practical concern, particularly in microscopy, medical imaging, and other fields producing large datasets.

However, limited storage does not automatically justify deleting source records that must be retained.

Researchers should consider institutional repositories, managed research storage, approved archival systems, and appropriate compression strategies.

Lossless compression may reduce storage requirements without discarding the information represented in the file, provided that relevant metadata and file integrity are preserved.

Lossy compression should not be assumed suitable for primary scientific images because it may alter pixel information.

When retention of all files is infeasible, decisions should follow an approved data-management plan and applicable obligations rather than informal deletion based on convenience.

Can Researchers Convert Proprietary Image Files Into Open Formats?

Potentially. Converting images into well-supported formats can improve long-term accessibility, particularly when proprietary software may become unavailable.

However, conversion must preserve the information needed for scientific interpretation.

For example, a native microscopy file may contain multiple channels, image planes, timestamps, spatial calibration, and acquisition metadata.

Exporting a single flattened image could discard much of that information.

Researchers should verify whether the chosen archival format preserves relevant pixel values, dimensions, metadata, and relationships among image components.

Where feasible, retaining the native acquisition file alongside a validated archival representation may provide additional protection against future software incompatibility.

How Can Researchers Protect Original Images From Accidental Modification?

Original files should be managed in a way that reduces the risk of unintended alteration.

A practical approach is to preserve a protected master copy and perform analysis on separate working copies.

Access controls, version management, and documented processing procedures can help maintain the distinction between original and derived data.

Cryptographic checksums, such as SHA-256 hashes, may be used to detect whether a file has changed since a recorded point in time.

However, a checksum establishes file consistency relative to a known value. It does not independently prove that the file was authentic when the checksum was created.

Researchers should therefore combine technical integrity checks with reliable acquisition and provenance records.

Is a Cloud Backup Enough to Preserve Scientific Images?

Cloud storage may be useful, but the existence of a synchronized copy does not necessarily constitute a reliable archival strategy.

For example, accidental deletion or corruption may propagate across synchronized devices.

Researchers should distinguish synchronization from backup and long-term archival preservation.

An appropriate storage strategy may include protected copies, access controls, version history, periodic integrity checks, and tested recovery procedures.

The exact arrangement should reflect institutional information-security requirements and the sensitivity of the data.

Where images contain identifiable participants or confidential information, approved storage environments should be used rather than personal or unauthorized cloud accounts.

What About Scientific Images Containing Personal or Sensitive Information?

Preservation must be balanced with privacy, consent, and legal requirements.

Medical images may contain identifiable metadata or features. Research photographs may show participants, workplaces, or other sensitive contexts.

Researchers should follow approved retention and access arrangements, including appropriate safeguards for identifiable data.

De-identification may be necessary for sharing, but the process should not inadvertently destroy information required for authorized scientific verification.

Where original identifiable files must be retained, access may need to be restricted while de-identified derivatives are used for analysis or publication.

Retention obligations do not authorize indefinite or unrestricted storage of personal information. Researchers must comply with applicable data-protection and ethics requirements.

Should Original Images Be Made Publicly Available?

Preserving original images and publicly sharing them are different obligations.

A journal or funder may require data availability, but access arrangements can vary according to privacy, consent, intellectual property, security, and other legitimate restrictions.

Some images may be deposited in public repositories. Others may require controlled access or retention within an institution.

Researchers should distinguish the obligation to maintain a verifiable research record from the separate question of who may access that record.

When data cannot be shared openly, the manuscript may need to explain the restrictions and any available access procedures according to the relevant publication requirements.

Can Original Images Help Resolve Research Integrity Allegations?

Yes. Original images and processing records may provide essential evidence when questions arise about duplicated panels, altered structures, inconsistent measurements, or unexpected similarities between figures.

For example, a journal may request unprocessed microscopy images to determine whether a published panel accurately represents the underlying observation.

Original files can help distinguish processing errors from more serious concerns, although their existence alone does not resolve every question.

The absence of original data may make an investigation more difficult, but it does not automatically establish that misconduct occurred.

Formal research misconduct determinations require evaluation under the applicable definitions and procedures.

What Happens if Original Scientific Images Have Already Been Lost?

Researchers should investigate whether authentic copies remain available in instrument storage, institutional backups, laboratory archives, collaborators' records, or other authorized locations.

If the original data cannot be recovered, the loss should be documented accurately.

Researchers should assess whether the remaining records are sufficient to verify the analysis and whether the loss affects any published or planned conclusions.

AI reconstruction should not be used to create replacement images and present them as recovered originals.

As discussed in AI completion of missing scientific image details, generated content may be plausible without reproducing the authentic observation.

Where the loss materially affects the reliability of reported findings, appropriate consultation with supervisors, research integrity offices, or journal editors may be necessary.

How Should AI Image Processing Be Documented Alongside the Originals?

When AI is used for image enhancement, reconstruction, segmentation, or measurement, the provenance record should identify the model and its role in the analytical workflow.

Relevant details may include the software and model version, processing settings, source image identifiers, generated outputs, and validation procedures.

Researchers should also record whether measurements were derived from original or processed images.

These details support the validation of AI-assisted scientific image analysis and help reviewers determine which aspects of the results depend on model-generated estimates.

Publication-specific requirements for disclosing AI image processing should be followed separately.

Preserving Original Microscopy Images During an AI-Assisted Analysis

Step 1: Identify the Essential Research Records

The researcher determines that the original acquisition files contain pixel information and metadata that are not fully preserved in the exported figures.

The segmentation masks and numerical measurements also depend on the original image characteristics.

Step 2: Establish a Provenance Structure

Each original image receives a stable identifier that links it to its processed versions, segmentation masks, and measurements.

The researcher records the denoising model, segmentation model, software versions, and relevant processing parameters.

Step 3: Separate Original and Derived Files

Step 4: Protect the Original Data

The researcher places the original files in an approved institutional storage environment with appropriate access controls and backup procedures.

Processing is performed on working copies rather than directly modifying the preserved originals.

Step 5: Verify That the Record Can Be Reconstructed

The researcher selects several published or analytically important results and traces them back through the measurements, masks, processed images, and original acquisitions.

This check helps identify missing links in the provenance record before the project is archived.

Step 6: Apply the Appropriate Retention Schedule

The researcher consults the institution's research-data policy and any relevant funder or publication requirements.

The original images and associated records are retained for the applicable period, rather than deleted merely because the manuscript has been completed.

The example illustrates that preservation concerns the integrity of the complete evidential chain, not simply maintaining a folder of final figures.

Common Misconceptions About Preserving Scientific Images

How to Establish a Defensible Scientific Image Preservation Workflow

Preservation should be planned before images are collected, particularly when the study involves substantial processing, AI reconstruction, or quantitative image analysis.

The goal is to maintain a trustworthy connection between the original observation and every result derived from it.

A Practical Prompt for Planning Scientific Image Preservation

AI can help organize a preservation plan, but institutional data stewards, research integrity offices, information-security personnel, and applicable policies should guide consequential retention and disposal decisions.

Before archiving the project, verify that a representative result can be traced from the published figure or numerical measurement back to the authentic source image and relevant processing records.

Before Archiving or Disposing of Scientific Images, Verify These

Frequently Asked Questions About Preserving Original Scientific Images

Preserve the Observation, Not Just Its Final Appearance

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.