Pre-Submission Review for Radiology Papers

Radiology reviewers usually ask:

• does the clinical use case matter to radiologists or patient care?
• are modality, protocol, scanner, acquisition, and image-selection details clear?
• is the unit of analysis patient-level, lesion-level, image-level, or study-level?
• are readers, adjudication, blinding, and reference standards described?
• are diagnostic accuracy claims reported with enough detail?
• if AI is involved, are training, validation, external testing, leakage controls, and demographic reporting clear?
• are figures readable without the author narrating them?
• does the target journal fit clinical radiology, AI, methods, education, intervention, or specialty imaging?

Those questions decide whether a paper feels reviewable.

In our pre-submission review work, radiology manuscripts most often fail because the manuscript does not make the imaging decision unit visible enough.

Patient-level versus lesion-level confusion: the abstract claims clinical diagnostic performance, but the analysis is organized around images, slices, lesions, or exams.

Reader-study gap: the manuscript reports model performance but does not explain whether radiologists read independently, with assistance, after washout, or against a clear reference standard.

Protocol opacity: scanner, sequence, reconstruction, contrast, timing, or inclusion details are not clear enough for another center to understand the dataset.

AI validation weakness: the model has internal validation but no external dataset, subgroup audit, calibration, or leakage analysis.

Figure trust problem: the figure is visually impressive but does not show the exact finding a skeptical reviewer needs to inspect.

European Radiology's submission guidance lists required manuscript files and article-type limits. Original articles are capped at 3,000 words, with structured abstracts, figure and table limits, and graphical abstracts required for Original Articles from 2025 onward. The journal also points authors to detailed manuscript requirements and disclosure language.

European Radiology Experimental points authors toward STARD for diagnostic accuracy, plus EQUATOR for study-type reporting. That matters because many radiology papers are diagnostic or prediction-model manuscripts, not generic clinical studies.

RSNA AI resources emphasize medical imaging AI reporting, including CLAIM. The CLAIM 2024 update is registered through EQUATOR and exists because AI imaging papers need complete reporting around data, model development, evaluation, and clinical relevance.

Send the manuscript, target journal, figures, tables, supplement, reporting checklist, reader-study protocol if relevant, data split description, ethics and consent language, device or AI-use disclosures if relevant, and prior decision letters if available.

For AI manuscripts, send model architecture, training and validation split details, external test information, cohort diagram, demographic table, calibration or error analysis, code/data availability statement, and any failure-case examples. For diagnostic accuracy studies, send reference standard details and patient-flow information.

A useful radiology pre-submission review should include:

• clinical use-case verdict
• imaging protocol clarity check
• reader-study and reference-standard critique
• diagnostic accuracy reporting risk
• AI validation and leakage critique if relevant
• figure readability and legend critique
• journal-lane fit
• submit, revise, retarget, or diagnose deeper call

The review should name the specific risk. "Figures need improvement" is too vague. A useful review says, "Figure 2 shows representative cases, but the paper needs false positives and false negatives because the clinical risk is overcalling low-prevalence disease."

Before submission, authors often need to:

• define the unit of analysis in the abstract and methods
• add a patient or exam flow diagram
• clarify reader selection, blinding, and adjudication
• add STARD, TRIPOD, CLAIM, CONSORT, STROBE, or PRISMA elements
• explain scanner, modality, acquisition, and reconstruction details
• move external validation or subgroup audit from supplement into the main story
• replace decorative examples with reviewer-useful examples
• retarget from a flagship radiology journal to AI, methods, specialty, or clinical imaging venue

These are readiness fixes, not line edits.

When a radiology manuscript has multiple risks, fix the layer that controls reviewer trust.

1. Clinical use case: explain why this imaging decision matters.
2. Analysis unit: make patient, lesion, exam, image, and reader units unmistakable.
3. Reference standard and reader design: show what the comparison actually means.
4. Reporting completeness: add the relevant checklist and flow details.
5. Figure readability: make figures interpretable at reviewer speed.

That order avoids the common mistake of making figures prettier while the diagnostic claim remains unstable.

A radiology editor is reading for use case and trust before polish. The first page should make the imaging problem, patient or exam unit, reference standard, and target reader visible without forcing the editor to reconstruct the study from the supplement.

If the abstract reports accuracy but does not say whether performance is patient-level or lesion-level, the editor expects reviewer friction. If an AI paper claims clinical readiness without external validation or clear reader comparison, the editor will usually see a development paper, not a deployment paper. If representative images are beautiful but the error cases are hidden, the submission can feel less honest than the data deserve.

For radiology manuscripts, first-page trust comes from showing how the image evidence maps to a real clinical decision.