Why English? | Reading CLEAR Radiomics Research Through Imaging, Segmentation, Features, Validation and Open Science

ORIENTATION

What this guide owns

Current fact check

The 2023 CLEAR publication presents 58 minimum reporting items, of which 56 are essential and two are recommended. It also presents the 43-item CLEAR-S shortened checklist. The guideline covers clinical radiomics research as a whole report.

Canonical boundary

This article owns reporting of clinical radiomics research with CLEAR. CLAIM 2024 owns broader medical-imaging AI reporting, PRIME 2.0 owns cardiovascular-imaging AI, and clinical utility, regulation and implementation require separate evidence.

Worked example

Every chapter returns to a fictional multicentre MRI radiomics study predicting which knee injuries are likely to need specialist review. It is fictional, low-stakes and educational; it does not report a real school, clinical decision or guaranteed outcome.

Warm up with eduKate’s guides to explaining scientific evidence and asking better questions in class.

CHAPTER 1 OF 16

1. Identify the Radiomics Method in the Title

Back to contents

Why this move matters

State radiomics, modality, task and important design features such as multicentre or external validation where space permits.

Follow the image-to-claim pipeline

State radiomics, modality, task and important design features such as multicentre or external validation where space permits. Move in order from acquisition and reconstruction to segmentation, feature extraction, reduction, modelling, validation and interpretation.

Protect every data boundary

Worked example: The title names MRI radiomics, prediction, multicentre data and external testing. Mark patient, examination, image, region, feature and outcome identifiers so training, tuning and testing remain genuinely separate.

Ask whether the signal survives change

Check scanners, protocols, reconstruction, segmentation, preprocessing, feature definitions, sites and time. A radiomic signature is useful only if its meaning is not an accident of one pipeline.

Keep prediction distinct from clinical benefit

A detailed title cannot compensate for hidden methods. CLEAR improves reporting of radiomics research; it does not turn association into causation, accuracy into usefulness or a retrospective model into a clinical tool.

Student, parent and career transfer

Make one evidence card for “Identify the Radiomics Method in the Title”: source location, plain-language meaning, limit and next question. A parent or teacher can ask for one everyday explanation and one specialist explanation; both must preserve the same conditions. This habit supports radiology, medical imaging, data science, biomedical engineering, clinical research, journal review, health technology and science communication without promising a grade, admission, funding decision or job.

CHAPTER 2 OF 16

2. Define the Clinical Question and Intended Use

Back to contents

Why this move matters

Describe the problem, current practice, knowledge gap, target population and how a radiomic output might affect a decision.

Follow the image-to-claim pipeline

Describe the problem, current practice, knowledge gap, target population and how a radiomic output might affect a decision. Move in order from acquisition and reconstruction to segmentation, feature extraction, reduction, modelling, validation and interpretation.

Protect every data boundary

Worked example: The model is intended to support referral prioritisation, not replace diagnosis. Mark patient, examination, image, region, feature and outcome identifiers so training, tuning and testing remain genuinely separate.

Ask whether the signal survives change

Check scanners, protocols, reconstruction, segmentation, preprocessing, feature definitions, sites and time. A radiomic signature is useful only if its meaning is not an accident of one pipeline.

Keep prediction distinct from clinical benefit

A technically feasible prediction may not answer a useful clinical question. CLEAR improves reporting of radiomics research; it does not turn association into causation, accuracy into usefulness or a retrospective model into a clinical tool.

Student, parent and career transfer

Make one evidence card for “Define the Clinical Question and Intended Use”: source location, plain-language meaning, limit and next question. A parent or teacher can ask for one everyday explanation and one specialist explanation; both must preserve the same conditions. This habit supports radiology, medical imaging, data science, biomedical engineering, clinical research, journal review, health technology and science communication without promising a grade, admission, funding decision or job.

CHAPTER 3 OF 16

3. Describe Cohort Assembly and Data Sources

Back to contents

Why this move matters

Report eligibility, recruitment dates, institutions, retrospective or prospective design, exclusions, participant flow and public or institutional sources.

Follow the image-to-claim pipeline

Report eligibility, recruitment dates, institutions, retrospective or prospective design, exclusions, participant flow and public or institutional sources. Move in order from acquisition and reconstruction to segmentation, feature extraction, reduction, modelling, validation and interpretation.

Protect every data boundary

Worked example: Each hospital’s eligible examinations and exclusions are shown separately. Mark patient, examination, image, region, feature and outcome identifiers so training, tuning and testing remain genuinely separate.

Ask whether the signal survives change

Check scanners, protocols, reconstruction, segmentation, preprocessing, feature definitions, sites and time. A radiomic signature is useful only if its meaning is not an accident of one pipeline.

Keep prediction distinct from clinical benefit

An image count can conceal repeated scans from the same person. CLEAR improves reporting of radiomics research; it does not turn association into causation, accuracy into usefulness or a retrospective model into a clinical tool.

Student, parent and career transfer

Make one evidence card for “Describe Cohort Assembly and Data Sources”: source location, plain-language meaning, limit and next question. A parent or teacher can ask for one everyday explanation and one specialist explanation; both must preserve the same conditions. This habit supports radiology, medical imaging, data science, biomedical engineering, clinical research, journal review, health technology and science communication without promising a grade, admission, funding decision or job.

CHAPTER 4 OF 16

4. Define Outcomes and Reference Standards

Back to contents

Why this move matters

Explain target labels, assessors, timing, blinding, disagreements and the relationship between imaging and outcome measurement.

Follow the image-to-claim pipeline

Explain target labels, assessors, timing, blinding, disagreements and the relationship between imaging and outcome measurement. Move in order from acquisition and reconstruction to segmentation, feature extraction, reduction, modelling, validation and interpretation.

Protect every data boundary

Worked example: Specialist review within a defined period supplies the reference outcome. Mark patient, examination, image, region, feature and outcome identifiers so training, tuning and testing remain genuinely separate.

Ask whether the signal survives change

Check scanners, protocols, reconstruction, segmentation, preprocessing, feature definitions, sites and time. A radiomic signature is useful only if its meaning is not an accident of one pipeline.

Keep prediction distinct from clinical benefit

A noisy or subjective label limits the model no matter how complex it is. CLEAR improves reporting of radiomics research; it does not turn association into causation, accuracy into usefulness or a retrospective model into a clinical tool.

Student, parent and career transfer

Make one evidence card for “Define Outcomes and Reference Standards”: source location, plain-language meaning, limit and next question. A parent or teacher can ask for one everyday explanation and one specialist explanation; both must preserve the same conditions. This habit supports radiology, medical imaging, data science, biomedical engineering, clinical research, journal review, health technology and science communication without promising a grade, admission, funding decision or job.

CHAPTER 5 OF 16

5. Report Image Acquisition and Reconstruction

Back to contents

Why this move matters

Give modality, scanner, vendor, field strength, sequences, parameters, contrast, reconstruction and protocol variation across sites.

Follow the image-to-claim pipeline

Give modality, scanner, vendor, field strength, sequences, parameters, contrast, reconstruction and protocol variation across sites. Move in order from acquisition and reconstruction to segmentation, feature extraction, reduction, modelling, validation and interpretation.

Protect every data boundary

Worked example: The MRI acquisition table separates scanner models and sequences by hospital. Mark patient, examination, image, region, feature and outcome identifiers so training, tuning and testing remain genuinely separate.

Ask whether the signal survives change

Check scanners, protocols, reconstruction, segmentation, preprocessing, feature definitions, sites and time. A radiomic signature is useful only if its meaning is not an accident of one pipeline.

Keep prediction distinct from clinical benefit

A feature can encode acquisition differences instead of biology. CLEAR improves reporting of radiomics research; it does not turn association into causation, accuracy into usefulness or a retrospective model into a clinical tool.

Student, parent and career transfer

Make one evidence card for “Report Image Acquisition and Reconstruction”: source location, plain-language meaning, limit and next question. A parent or teacher can ask for one everyday explanation and one specialist explanation; both must preserve the same conditions. This habit supports radiology, medical imaging, data science, biomedical engineering, clinical research, journal review, health technology and science communication without promising a grade, admission, funding decision or job.

CHAPTER 6 OF 16

6. Describe Image Quality Control

Back to contents

Why this move matters

State checks, exclusions, artefact handling, reviewer roles and whether quality decisions occurred before outcomes were known.

Follow the image-to-claim pipeline

State checks, exclusions, artefact handling, reviewer roles and whether quality decisions occurred before outcomes were known. Move in order from acquisition and reconstruction to segmentation, feature extraction, reduction, modelling, validation and interpretation.

Protect every data boundary

Worked example: Motion artefacts are graded using a prespecified rule by blinded reviewers. Mark patient, examination, image, region, feature and outcome identifiers so training, tuning and testing remain genuinely separate.

Ask whether the signal survives change

Check scanners, protocols, reconstruction, segmentation, preprocessing, feature definitions, sites and time. A radiomic signature is useful only if its meaning is not an accident of one pipeline.

Keep prediction distinct from clinical benefit

Subjective quality removal can introduce selection bias. CLEAR improves reporting of radiomics research; it does not turn association into causation, accuracy into usefulness or a retrospective model into a clinical tool.

Student, parent and career transfer

Make one evidence card for “Describe Image Quality Control”: source location, plain-language meaning, limit and next question. A parent or teacher can ask for one everyday explanation and one specialist explanation; both must preserve the same conditions. This habit supports radiology, medical imaging, data science, biomedical engineering, clinical research, journal review, health technology and science communication without promising a grade, admission, funding decision or job.

CHAPTER 7 OF 16

7. Open the Segmentation Process

Back to contents

Why this move matters

Report region definition, manual or automated method, software, raters, blinding, training, repeatability and disagreement.

Follow the image-to-claim pipeline

Report region definition, manual or automated method, software, raters, blinding, training, repeatability and disagreement. Move in order from acquisition and reconstruction to segmentation, feature extraction, reduction, modelling, validation and interpretation.

Protect every data boundary

Worked example: Two readers segment injuries on a subset to assess robustness. Mark patient, examination, image, region, feature and outcome identifiers so training, tuning and testing remain genuinely separate.

Ask whether the signal survives change

Check scanners, protocols, reconstruction, segmentation, preprocessing, feature definitions, sites and time. A radiomic signature is useful only if its meaning is not an accident of one pipeline.

Keep prediction distinct from clinical benefit

A region boundary can strongly change extracted features. CLEAR improves reporting of radiomics research; it does not turn association into causation, accuracy into usefulness or a retrospective model into a clinical tool.

Student, parent and career transfer

Make one evidence card for “Open the Segmentation Process”: source location, plain-language meaning, limit and next question. A parent or teacher can ask for one everyday explanation and one specialist explanation; both must preserve the same conditions. This habit supports radiology, medical imaging, data science, biomedical engineering, clinical research, journal review, health technology and science communication without promising a grade, admission, funding decision or job.

CHAPTER 8 OF 16

8. Report Preprocessing and Harmonisation

Back to contents

Why this move matters

Describe resampling, normalisation, discretisation, filtering, registration and harmonisation, with parameters and order.

Follow the image-to-claim pipeline

Describe resampling, normalisation, discretisation, filtering, registration and harmonisation, with parameters and order. Move in order from acquisition and reconstruction to segmentation, feature extraction, reduction, modelling, validation and interpretation.

Protect every data boundary

Worked example: Voxel size is standardised before feature calculation and the exact settings are reported. Mark patient, examination, image, region, feature and outcome identifiers so training, tuning and testing remain genuinely separate.

Ask whether the signal survives change

Check scanners, protocols, reconstruction, segmentation, preprocessing, feature definitions, sites and time. A radiomic signature is useful only if its meaning is not an accident of one pipeline.

Keep prediction distinct from clinical benefit

Changing preprocessing order can change the feature values. CLEAR improves reporting of radiomics research; it does not turn association into causation, accuracy into usefulness or a retrospective model into a clinical tool.

Student, parent and career transfer

Make one evidence card for “Report Preprocessing and Harmonisation”: source location, plain-language meaning, limit and next question. A parent or teacher can ask for one everyday explanation and one specialist explanation; both must preserve the same conditions. This habit supports radiology, medical imaging, data science, biomedical engineering, clinical research, journal review, health technology and science communication without promising a grade, admission, funding decision or job.

CHAPTER 9 OF 16

9. Define Feature Extraction and Software

Back to contents

Why this move matters

Name feature families, mathematical definitions or standards, software, versions, settings and deep-feature architecture where relevant.

Follow the image-to-claim pipeline

Name feature families, mathematical definitions or standards, software, versions, settings and deep-feature architecture where relevant. Move in order from acquisition and reconstruction to segmentation, feature extraction, reduction, modelling, validation and interpretation.

Protect every data boundary

Worked example: The report cites standard feature definitions and the exact extraction library version. Mark patient, examination, image, region, feature and outcome identifiers so training, tuning and testing remain genuinely separate.

Ask whether the signal survives change

Check scanners, protocols, reconstruction, segmentation, preprocessing, feature definitions, sites and time. A radiomic signature is useful only if its meaning is not an accident of one pipeline.

Keep prediction distinct from clinical benefit

Feature names can differ across software implementations. CLEAR improves reporting of radiomics research; it does not turn association into causation, accuracy into usefulness or a retrospective model into a clinical tool.

Student, parent and career transfer

Make one evidence card for “Define Feature Extraction and Software”: source location, plain-language meaning, limit and next question. A parent or teacher can ask for one everyday explanation and one specialist explanation; both must preserve the same conditions. This habit supports radiology, medical imaging, data science, biomedical engineering, clinical research, journal review, health technology and science communication without promising a grade, admission, funding decision or job.

CHAPTER 10 OF 16

10. Test Feature Robustness and Reduce Dimensions

Back to contents

Why this move matters

Explain repeatability, stability across perturbations, collinearity, feature selection and dimensionality reduction within the correct data partition.

Follow the image-to-claim pipeline

Explain repeatability, stability across perturbations, collinearity, feature selection and dimensionality reduction within the correct data partition. Move in order from acquisition and reconstruction to segmentation, feature extraction, reduction, modelling, validation and interpretation.

Protect every data boundary

Worked example: Unstable features are removed using training data only. Mark patient, examination, image, region, feature and outcome identifiers so training, tuning and testing remain genuinely separate.

Ask whether the signal survives change

Check scanners, protocols, reconstruction, segmentation, preprocessing, feature definitions, sites and time. A radiomic signature is useful only if its meaning is not an accident of one pipeline.

Keep prediction distinct from clinical benefit

Selecting features before the split leaks test information. CLEAR improves reporting of radiomics research; it does not turn association into causation, accuracy into usefulness or a retrospective model into a clinical tool.

Student, parent and career transfer

Make one evidence card for “Test Feature Robustness and Reduce Dimensions”: source location, plain-language meaning, limit and next question. A parent or teacher can ask for one everyday explanation and one specialist explanation; both must preserve the same conditions. This habit supports radiology, medical imaging, data science, biomedical engineering, clinical research, journal review, health technology and science communication without promising a grade, admission, funding decision or job.

CHAPTER 11 OF 16

11. Protect Training, Tuning and Testing Boundaries

Back to contents

Why this move matters

Report split units, site or time separation, patient grouping, cross-validation, nested procedures and when each dataset was accessed.

Follow the image-to-claim pipeline

Report split units, site or time separation, patient grouping, cross-validation, nested procedures and when each dataset was accessed. Move in order from acquisition and reconstruction to segmentation, feature extraction, reduction, modelling, validation and interpretation.

Protect every data boundary

Worked example: All examinations from one patient stay in one partition, and one hospital is held out. Mark patient, examination, image, region, feature and outcome identifiers so training, tuning and testing remain genuinely separate.

Ask whether the signal survives change

Check scanners, protocols, reconstruction, segmentation, preprocessing, feature definitions, sites and time. A radiomic signature is useful only if its meaning is not an accident of one pipeline.

Keep prediction distinct from clinical benefit

Random image splits can place near-duplicates across training and test sets. CLEAR improves reporting of radiomics research; it does not turn association into causation, accuracy into usefulness or a retrospective model into a clinical tool.

Student, parent and career transfer

Make one evidence card for “Protect Training, Tuning and Testing Boundaries”: source location, plain-language meaning, limit and next question. A parent or teacher can ask for one everyday explanation and one specialist explanation; both must preserve the same conditions. This habit supports radiology, medical imaging, data science, biomedical engineering, clinical research, journal review, health technology and science communication without promising a grade, admission, funding decision or job.

CHAPTER 12 OF 16

12. Describe Models, Baselines and Hyperparameters

Back to contents

Why this move matters

Name algorithms, inputs, tuning ranges, optimisation, stopping, random seeds and simpler or clinical comparators.

Follow the image-to-claim pipeline

Name algorithms, inputs, tuning ranges, optimisation, stopping, random seeds and simpler or clinical comparators. Move in order from acquisition and reconstruction to segmentation, feature extraction, reduction, modelling, validation and interpretation.

Protect every data boundary

Worked example: The radiomics model is compared with a clinical baseline and a combined model. Mark patient, examination, image, region, feature and outcome identifiers so training, tuning and testing remain genuinely separate.

Ask whether the signal survives change

Check scanners, protocols, reconstruction, segmentation, preprocessing, feature definitions, sites and time. A radiomic signature is useful only if its meaning is not an accident of one pipeline.

Keep prediction distinct from clinical benefit

Testing many models and reporting only the best inflates optimism. CLEAR improves reporting of radiomics research; it does not turn association into causation, accuracy into usefulness or a retrospective model into a clinical tool.

Student, parent and career transfer

Make one evidence card for “Describe Models, Baselines and Hyperparameters”: source location, plain-language meaning, limit and next question. A parent or teacher can ask for one everyday explanation and one specialist explanation; both must preserve the same conditions. This habit supports radiology, medical imaging, data science, biomedical engineering, clinical research, journal review, health technology and science communication without promising a grade, admission, funding decision or job.

CHAPTER 13 OF 16

13. Report Performance With Uncertainty and Calibration

Back to contents

Why this move matters

Use task-appropriate metrics, confidence intervals, calibration, thresholds, class balance and error analysis rather than one headline score.

Follow the image-to-claim pipeline

Use task-appropriate metrics, confidence intervals, calibration, thresholds, class balance and error analysis rather than one headline score. Move in order from acquisition and reconstruction to segmentation, feature extraction, reduction, modelling, validation and interpretation.

Protect every data boundary

Worked example: Discrimination, calibration and errors are reported on the external site. Mark patient, examination, image, region, feature and outcome identifiers so training, tuning and testing remain genuinely separate.

Ask whether the signal survives change

Check scanners, protocols, reconstruction, segmentation, preprocessing, feature definitions, sites and time. A radiomic signature is useful only if its meaning is not an accident of one pipeline.

Keep prediction distinct from clinical benefit

A high area under the curve does not identify a useful decision threshold. CLEAR improves reporting of radiomics research; it does not turn association into causation, accuracy into usefulness or a retrospective model into a clinical tool.

Student, parent and career transfer

Make one evidence card for “Report Performance With Uncertainty and Calibration”: source location, plain-language meaning, limit and next question. A parent or teacher can ask for one everyday explanation and one specialist explanation; both must preserve the same conditions. This habit supports radiology, medical imaging, data science, biomedical engineering, clinical research, journal review, health technology and science communication without promising a grade, admission, funding decision or job.

CHAPTER 14 OF 16

14. Use Unseen External Testing and Subgroup Checks

Back to contents

Why this move matters

Describe genuinely unseen data, site differences, subgroup performance and transport limits.

Follow the image-to-claim pipeline

Describe genuinely unseen data, site differences, subgroup performance and transport limits. Move in order from acquisition and reconstruction to segmentation, feature extraction, reduction, modelling, validation and interpretation.

Protect every data boundary

Worked example: The final model is locked before testing at the fourth hospital. Mark patient, examination, image, region, feature and outcome identifiers so training, tuning and testing remain genuinely separate.

Ask whether the signal survives change

Check scanners, protocols, reconstruction, segmentation, preprocessing, feature definitions, sites and time. A radiomic signature is useful only if its meaning is not an accident of one pipeline.

Keep prediction distinct from clinical benefit

One external site is useful but not universal validation. CLEAR improves reporting of radiomics research; it does not turn association into causation, accuracy into usefulness or a retrospective model into a clinical tool.

Student, parent and career transfer

Make one evidence card for “Use Unseen External Testing and Subgroup Checks”: source location, plain-language meaning, limit and next question. A parent or teacher can ask for one everyday explanation and one specialist explanation; both must preserve the same conditions. This habit supports radiology, medical imaging, data science, biomedical engineering, clinical research, journal review, health technology and science communication without promising a grade, admission, funding decision or job.

CHAPTER 15 OF 16

15. Interpret Biology and Clinical Utility Cautiously

Back to contents

Why this move matters

Separate association, prediction and causation; discuss workflow, decision consequences, errors, benefit and the evidence still needed.

Follow the image-to-claim pipeline

Separate association, prediction and causation; discuss workflow, decision consequences, errors, benefit and the evidence still needed. Move in order from acquisition and reconstruction to segmentation, feature extraction, reduction, modelling, validation and interpretation.

Protect every data boundary

Worked example: The result is presented as a referral-support hypothesis requiring prospective evaluation. Mark patient, examination, image, region, feature and outcome identifiers so training, tuning and testing remain genuinely separate.

Ask whether the signal survives change

Check scanners, protocols, reconstruction, segmentation, preprocessing, feature definitions, sites and time. A radiomic signature is useful only if its meaning is not an accident of one pipeline.

Keep prediction distinct from clinical benefit

Feature importance does not prove a biological mechanism. CLEAR improves reporting of radiomics research; it does not turn association into causation, accuracy into usefulness or a retrospective model into a clinical tool.

Student, parent and career transfer

Make one evidence card for “Interpret Biology and Clinical Utility Cautiously”: source location, plain-language meaning, limit and next question. A parent or teacher can ask for one everyday explanation and one specialist explanation; both must preserve the same conditions. This habit supports radiology, medical imaging, data science, biomedical engineering, clinical research, journal review, health technology and science communication without promising a grade, admission, funding decision or job.

CHAPTER 16 OF 16

16. Enable Reproduction and State Limits

Back to contents

Why this move matters

Report protocols, code, data access, models, feature definitions, funding, conflicts, limitations and barriers to sharing.

Follow the image-to-claim pipeline

Report protocols, code, data access, models, feature definitions, funding, conflicts, limitations and barriers to sharing. Move in order from acquisition and reconstruction to segmentation, feature extraction, reduction, modelling, validation and interpretation.

Protect every data boundary

Worked example: A versioned pipeline and data dictionary are shared while protected images follow controlled access. Mark patient, examination, image, region, feature and outcome identifiers so training, tuning and testing remain genuinely separate.

Ask whether the signal survives change

Check scanners, protocols, reconstruction, segmentation, preprocessing, feature definitions, sites and time. A radiomic signature is useful only if its meaning is not an accident of one pipeline.

Keep prediction distinct from clinical benefit

Open code cannot repair biased sampling or weak labels. CLEAR improves reporting of radiomics research; it does not turn association into causation, accuracy into usefulness or a retrospective model into a clinical tool.

Student, parent and career transfer

Make one evidence card for “Enable Reproduction and State Limits”: source location, plain-language meaning, limit and next question. A parent or teacher can ask for one everyday explanation and one specialist explanation; both must preserve the same conditions. This habit supports radiology, medical imaging, data science, biomedical engineering, clinical research, journal review, health technology and science communication without promising a grade, admission, funding decision or job.

PRACTICE LAB

Build a clear decision trail

Student route

1. Choose one original report inside this guide’s scope.
2. Copy its objective and underline actor, action, context, evidence and intended decision.
3. Draw the specific reasoning chain used in this article.
4. Mark each step reported, inferred, contested or unknown.
5. Rewrite the conclusion so it preserves the narrowest important limitation.

Parent or teacher route

Ask the learner to explain follow a radiomics claim from medical image acquisition to a tested model without losing the pipeline that created the signal twice: once to a classmate and once to a method specialist. Praise a precise correction and a well-stated unknown; both show disciplined attention to evidence.

Early-career route

Create a one-page clear audit with source link, date checked, scope, method logic, decisive wording, contrary evidence, limit and next action. A colleague should be able to continue without guessing.

Continue with eduKate’s guides to reading course and career information and giving useful feedback in group projects.

USEFUL NEXT READING

Connect this narrow owner to its neighbour

Read the guide to explaining scientific evidence for a related but different reader job. Return here when the question is specifically how to follow a radiomics claim from medical image acquisition to a tested model without losing the pipeline that created the signal. This boundary keeps one canonical owner per intent.

For English practice across research, school and career communication, visit the English at work and in public life guide.

FREQUENTLY ASKED QUESTIONS

Questions students and parents often ask

What does CLEAR stand for?

CheckList for EvaluAtion of Radiomics research.

How many items are in CLEAR?

Fifty-eight: 56 essential and two recommended.

What is CLEAR-S?

A shortened 43-item checklist focused on methodological-quality items.

Why can image splitting cause leakage?

Images from the same patient or related acquisition can enter both training and testing, making the test less independent.

Does external validation prove clinical benefit?

No. It tests transport of performance; benefit needs clinical evaluation.

Why report software versions?

Feature definitions and calculations can vary across implementations and releases.