Judge the effect on scientific meaning
Research images are data or close representations of data, not merely decoration. Brightness, contrast, colour maps, denoising and cropping can change visible structures. A technically small adjustment may therefore have a material scholarly effect. Ask whether a knowledgeable reviewer with the raw file and log would see the same relevant evidence.
Fields differ. Microscopy, gels, clinical images, photographs and reconstructed measurement maps have their own conventions. Read the journal and field guidance before processing. Availability in editing software is not an ethical permission. This page offers general publication guidance rather than a forensic conclusion.
Keep original acquisition files unchanged and access-controlled. Generate manuscript images as new objects through logged steps. File identity, software release, parameters and responsible person belong in the audit trail. A screenshot is not a substitute for raw data.
Global processing can still mislead
A global tonal adjustment may be acceptable if every pixel is treated equally and faint signals are not clipped or bright signals saturated. Inspect the histogram and original side by side. Comparisons need the same acquisition and display conditions or a stated scientific reason. Unequal display ranges can create an apparent group difference.
Local adjustment is particularly risky. If an established method requires local correction, define it reproducibly and apply it under the same rule to all comparable images. Painting brightness onto a desired region is not neutral enhancement.
Denoising, sharpening and deconvolution can remove or create structures. Record algorithm, software version and settings, retain intermediate output and evaluate the method with suitable controls. An attractive processed image must not replace quantitative measurement.
Expose crops, duplicates and assembled panels
A crop directs attention but must preserve relevant context. Store coordinates or a full-frame companion. Removing a control area outside the crop can distort interpretation. Calculate scale bars from acquisition metadata rather than copying them by eye.
Panels from different experiments need visible divisions and unambiguous labels. Gel lanes must not be moved together as though they were adjacent in one untouched gel. Reusing a control is acceptable only where methodologically justified and transparently identified.
Accidental duplication often begins with poor file naming. Maintain an inventory linking every panel to raw file, sample, acquisition and analysis. Mirroring, rotation or a new crop must not hide duplicate use. For overlays, identify channels, registration method and stacking order.
Build a reversible figure workflow
- Import immutable raw files and capture checksums.
- Link every panel ID to sample, time point and acquisition.
- Record software, version, operation and parameter.
- Process comparison images under one declared rule.
- Inspect borders, labels, scale bars and legends.
- Ask a second person to trace each panel to raw evidence.
- Version the final figure with its processing log.
Review automated exports too. Presentation software can compress pixels, crop objects or convert colour profiles. Inspect the final PDF at high magnification and in an accessibility-appropriate view. The legend reports consequential processing and explains any discontinuity.
Quantitative analysis should ordinarily operate on defined data rather than a publication screenshot. If segmentation or measurement uses a processed image, identify that processing release and validate its effect. Keep display processing separate from analytical transformation.
Preserve originals when a concern is raised
Do not modify files after an enquiry. Secure raw images, submitted manuscript, processing log and software environment. Describe the precise anomaly and invite the responsible authors to supply originals and an explanation. Forensic tools generate leads; compression and export can create superficially similar patterns.
Journal and institution have distinct roles. The journal protects the publication record; an institution can examine provenance and possible conduct. Depending on reliability, a correction, expression of concern or retraction may follow. An anomalous image does not justify public presumption of intent.
Check whether the questioned image also supports data deposited elsewhere or a later publication. Related records are reviewed without assuming that one error contaminates everything. Preserve a clear chain from concern to decision and record any independent expert advice.
The post-publication process separates journal and institutional roles. The data availability template covers access to underlying material. More topics sit in the publication ethics hub. The one foundation route is the guide to evidence review.