radiology-deep-learning

Imaging Deep-Learning Study Design

Use this skill to design (or audit) an imaging deep-learning study so it is reproducible, honestly validated, and CLAIM-compliant. DL imaging papers get torn apart for slice-level splits, patient overlap, test-set tuning, no external validation, and baselines that are too weak to make "deep learning wins" mean anything. This skill encodes the architecture/training choices and the partition hygiene reviewers enforce.

Core stance

• Patient-level everything. Splits, augmentation, and any data-dependent step respect the patient boundary — slices/lesions/sequences/timepoints from one patient never span sets.
• Right capacity for the data. Small cohorts → transfer learning, self-supervised pretraining, strong simple baselines, heavy augmentation, and nested CV — not a giant model trained from scratch on 200 images.
• Beat a real baseline. "DL is better" needs a fair comparator: a radiomics/clinical model, a strong simpler network, or radiologists — tuned as carefully as the proposed model.
• Inputs declared. State exactly how images, masks, clinical variables, text, and molecular data enter the model (channels, crops, fusion point), and how missing modalities are handled.
• Validation must match the claim. Freeze the pipeline before evaluation. Internal patient-level resampling can support bounded development or feasibility claims; temporal, geographic, external, or deployment claims require a genuinely untouched population representing that transport step (→ radiology-design/validation-strategy).
• Report fit-for-purpose performance. Include uncertainty and failure cases. Use calibration for probabilistic predictions and decision-analytic utility only when a real decision and threshold range are defined (→ radiology-stats).
• Select trustworthiness analyses by risk and use. Interpretability, predictive uncertainty, OOD/robustness and fairness analyses are required when they answer a named failure mode, support a proposed use, or are part of a foundation/deployment evaluation; do not add them as ceremonial high-impact-journal checkboxes (→ interpretability-uncertainty.md).
• Integrity. Never invent performance, training curves, or hyperparameters; mark what must be run.

When to use

• "Design a CNN/Transformer/3D/segmentation/detection/prognostic imaging model." / "影像深度学习课题设计。"
• "Transfer learning vs self-supervised vs from scratch for my cohort size?"
• "How should images + clinical + pathology/text enter the model (multimodal fusion)?"
• "Augmentation, class imbalance, hyperparameter search, baselines — how to set up?"
• "Audit my DL Methods for slice-level leakage / patient overlap / test-set tuning."

When to open extra files

File	Open when
references/architecture-choice.md	Choosing 2D/2.5D/3D CNN, Transformer/ViT, segmentation/detection/prognostic heads; foundation models; capacity vs cohort size
references/training-protocol.md	Transfer/SSL/from-scratch, splits, augmentation, class imbalance, loss/optimizer/schedule, hyperparameter search, checkpointing, seeds
references/multimodal-inputs.md	How images/masks/clinical/text/molecular inputs enter the model; fusion strategies; missing-modality handling
references/dl-leakage-audit.md	The DL-specific leakage/validity checklist reviewers weaponise
references/interpretability-uncertainty.md	Explainability (Grad-CAM/SHAP/attention) reported without overclaiming; uncertainty quantification (MC dropout, ensembles, conformal prediction); robustness/OOD testing; FUTURE-AI framing
references/foundation-models-trustworthy-ai.md	Foundation models, ViT, SSL, VLM/report generation, 3D radiology models, adapter/LoRA, UQ/XAI/causal robustness, or deployment-grade trustworthy AI
references/foundation-model-data-genealogy-and-privacy.md	Pretraining sources are reused, closed or uncertain; audit patient/derivative overlap, contamination wording, memorisation/privacy threats, provider retention or federated leakage
references/experiment-and-model-card.md	Freezing experiment provenance, model/data cards, run manifests, negative results, test-set access, compute/environment, or a final reproducibility package

Workflow

1. Confirm the design (reuse radiology-design) — task, endpoint, unit (patient-level), cohorts, validation type, realistic capacity given n.
2. Choose architecture (architecture-choice.md) — dimension, family, task head; justify capacity vs cohort size; pick the baseline(s) to beat.
3. For foundation/trustworthy-AI designs, open foundation-models-trustworthy-ai.md and, when pretraining sources are reused, closed or uncertain, foundation-model-data-genealogy-and-privacy.md. Specify the overlap-evidence state, adaptation, baseline ladder, calibration, UQ, XAI stability, OOD/robustness, equity/fairness, privacy threat model, and shortcut/confounder audits.
4. Define inputs (multimodal-inputs.md) — channels/crops/fusion; missing-modality rule; leakage-safe use of masks and clinical/text/molecular data.
5. Set training (training-protocol.md) — transfer/SSL/scratch, patient-level splits, augmentation, imbalance handling, loss/optimizer/schedule, nested-CV hyperparameter search, seeds, checkpoint selection (on validation, never test).
6. Validate — use patient-level internal evaluation and add temporal/geographic/external evaluation when the intended claim requires that transport step; keep the pipeline frozen and report claim-relevant performance, uncertainty and failure cases (→ radiology-stats).
7. Run justified trustworthiness analyses (interpretability-uncertainty.md) — name the failure mode or decision each XAI/UQ/OOD/robustness analysis addresses, choose a method matched to the architecture, and report bounded results rather than only flattering examples.
8. Freeze experiment provenance (experiment-and-model-card.md) — data/split manifest, configurations, seeds, compute/environment, checkpoint rule, test-access log, model card, failure/negative-result log, and deviations.
9. Audit leakage (dl-leakage-audit.md) and write Methods to CLAIM 2024.

Output contract

1. Model design — architecture, task head, capacity rationale, baseline(s).
2. Input spec — how each modality enters; fusion point; missing-modality handling.
3. Training protocol — pretraining strategy, splits, augmentation, imbalance, loss/optim/ schedule, hyperparameter search, seeds, checkpoint rule — reproducibly.
4. Validation plan — internal evaluation plus any claim-required transport evaluation; fit-for-purpose metrics, uncertainty and failure cases.
5. Interpretability & uncertainty — when justified, the named risk/use, method, parameters, quantitative check, and bounded claim it supports (→ interpretability-uncertainty.md).
6. Trustworthiness modules — for foundation/VLM/deployment-grade models: calibration, UQ, OOD/fairness, XAI stability, shortcut/confounder audit, pretraining-overlap state and a use-specific privacy threat model; do not infer fairness from subgroup AUC alone.
7. Leakage audit — pass/fail per item + fix.
8. Experiment/model record — data/split/run manifests, model card, configuration and checkpoint provenance, environment/compute, negative results, and deviations actually available.
9. Table handoff — baseline/ablation/performance/calibration/fairness value keys and configuration rows for radiology-table.
10. Methods paragraph — CLAIM-aligned prose (+ 待确认 for Chinese authors).

Quality bar

A good DL design is reproducible from the protocol, splits at the patient level, beats a fair baseline, freezes the pipeline before claim-matched evaluation, and reports relevant uncertainty and failure cases — not a single AUC from a slice-level split.

Handoffs

• Series/phase/sequence qualification, acquisition/reconstruction, quantitative transforms, artifacts, dose, phantom/test-retest and protocol drift → radiology-acquisition-qc; consume its accepted-image gate and deviations before training/evaluation.
• Hand-crafted feature comparison / deep-feature extraction context → radiology-radiomics.
• Segmentation/detection ground-truth mask SOP and reader reproducibility → radiology-annotation.
• CLAIM/TRIPOD+AI audit, FUTURE-AI/TRIPOD-LLM edge cases → radiology-reporting.
• Metrics, CIs, DeLong, calibration, MRMC, sample size → radiology-stats.
• Hyperparameter/configuration sensitivity, matched benchmark fairness, component ablation, seed/run failure boundaries and efficiency evidence → radiology-method-evaluation; this skill still owns model/training implementation.
• Validation-type design (external/temporal/multi-center) → radiology-design.
• Biological interpretation of deep features → radiology-radiogenomics.
• Reader study / prospective deployment / monitoring for drift → radiology-translation.
• Figures (architecture, ROC, calibration, Grad-CAM, uncertainty plots) → radiology-figure.
• Tables (performance, ablation, subgroup/fairness, hyperparameters, failure cases) → radiology-table.
• Reframing this as a funding proposal instead of / alongside a paper → radiology-grant.
• Full-project state, provenance, and cross-artifact consistency → radiology-pipeline.