Medical Imaging AI Literature Review Skill (v3.2.0)

Produce comprehensive reviews that pass first-round peer review on factual grounds, not just structural grounds.

This is not a template-filling skill. It is a write-with-verify discipline.

Quick Start

First choose the review type. Read references/REVIEW_TYPES.md before collecting literature or drafting prose.

Default narrative/method-survey projects live in 4 files:

project_root/
├── PARADIGM.md            # Style spec from 2-3 exemplar reviews (Phase 0)
├── CLAUDE.md              # Project-specific terminology + literature inventory
├── IMPLEMENTATION_PLAN.md # 3-axis outline + per-claim verification checklist
└── manuscript_draft.md    # The actual manuscript

Scoping and systematic reviews add protocol, search, screening, extraction, and risk-of-bias files; see references/REVIEW_TYPES.md and references/REPORTING_STANDARDS.md.

Follow the workflow in references/WORKFLOW.md. The phases are: review-type routing -> paradigm capture -> init -> collect-and-verify -> outline/taxonomy -> write-with-per-claim-verification -> peer review -> submission prep.

Review Type Routing

Do not let the title outrun the methods.

If the manuscript uses the phrase "systematic review", it must contain reproducible search strings, eligibility criteria, screening flow, extraction fields, and risk-of-bias methods. Otherwise, call it a narrative review, method survey, or scoping review.

Core Principles

Writing voice — match strength to evidence, not hedge by default

Calibrate language to evidence strength, not to a fixed hedging register.

When ≥2 independent peer-reviewed groups confirm a finding, state it strongly. When evidence is single-source or contested, state it cautiously. When evidence is absent, say so.

Avoid the LLM tells:

• "has shown promising results"
• "may suggest"
• "interestingly,"
• "it is worth noting that"
• "in recent years,"
• "demonstrates the effectiveness of"
• "may offer significant advantages"

These phrases are AI-detector top features. Real flagship-review authors don't use them. Strip them.

Take a position when evidence supports it. Neutral catalogue is the LLM default and the failure mode to avoid. See Verdict sentences below.

Citations — every claim verified before commit

Every [N] citation must satisfy these checks (the full 5-rule protocol is in references/CITATION_INTEGRITY.md):

1. The cited paper exists (DOI / PMID resolves on PubMed or Crossref) with no placeholder DOI.
2. The author list matches the first-source (especially first and last author).
3. The numeric claim in the body sentence (Dice, HR, sample size, etc.) appears in the cited paper's abstract or results section.
4. The directional claim in the body sentence (higher/lower, increased/decreased) matches the source's stated direction.
5. Clinical claims cite a peer-reviewed primary source, not a vendor white paper or regulatory letter.

If any check fails, the citation is broken — fix before continuing. This is a hard gate: even a single broken citation must be fixed before delivery.

Method descriptions — read first, write after

Do not fill in a template like [Author] et al. [ref] proposed [method]... Achieves Dice of X.XX. That template is a hallucination trap.

Use this discipline instead:

1. Read the actual paper (abstract + methods + results). Use whatever first-source route is available: PubMed/DOI pages, arXiv pages or PDFs, Zotero full text, local PDFs, institutional copies, or journal pages. Confirm available tools before assuming a specific MCP name.
2. Note the actual module names, the actual benchmark, the actual numbers, in your own working notes — not in the manuscript yet.
3. Write the method description from those notes, citing specific numbers and module names verbatim from the paper.
4. Verify by spot-checking 1-2 of the numbers against the paper one more time before moving on.

If you can't access the paper, do not write about its internal architecture or specific performance numbers — and do not assert priority or novelty ("first to", "首个", "earliest", "novel"). Priority claims are strong, falsifiable, and frequently wrong; asserting one you haven't verified is a hallucination. Instead, cite it for a neutral, non-priority contribution ("applied X to Y") and, if useful, note the claim is unverified — or leave it out.

Heading depth — match the target article type

• H2 (##) for top-level sections (Introduction, Methods, Applications, Discussion, ...).
• H3 (###) for subsections.
• In flagship narrative reviews, avoid H4 in body; use bold lead-in **Topic.** paragraph starters for deeper grouping.
• In systematic/scoping reviews, method subheadings may follow journal or PRISMA conventions even if that creates a more formal Methods section.
• Avoid number prefixes (1., 1.1, 1.2.3) unless the target journal explicitly requires numbered sections.

Equations — in a Box, not in body

Display equations (DSC, IoU, clDice, FedAvg, GCN propagation, ...) appear in Boxes, not inline in body paragraphs. Textbook formulas can be referenced ("the Dice similarity coefficient — see Box 1") but should not be displayed inline.

If a formula has no methodological insight worth displaying (e.g., FedAvg averaging), describe it in prose instead of showing it.

Vendor names — table-first, sparing in prose

Vendor names (HeartFlow, Cleerly, Caristo, Keya, Shukun, ...) belong primarily in the Commercial Products / Regulatory & Validation table. In body text use category descriptors unless the product name is necessary to define a regulatory fact, trial population, or head-to-head distinction.

• ✗ "HeartFlow's CT-FFR product was validated in NXT, ADVANCE, and PACIFIC..."
• ✓ "The first FDA-cleared CT-FFR product (Table N, row 1) was validated in NXT, ADVANCE, and PACIFIC..."
• ✓ "The table lists HeartFlow, Cleerly, Caristo, and other products with their regulatory status and peer-reviewed validation evidence."

Reason: repeated product names in body text read like marketing copy. Use exact product names when precision matters; cite peer-reviewed evidence for clinical claims.

Default Narrative / Method Survey Structure

# [Title]: <evocative subtitle>

## Key Points
- 4-5 bullets, each 1-3 sentences, expressing the main conclusions.

## Abstract

## Introduction
### Clinical background
### Technical challenge
### Scope and contributions

## Datasets and evaluation metrics
(Table 1: public datasets)
(Box 1: evaluation metrics with equations)

## Methods                              # 3-axis grouping is the default for method surveys
### Architectural priors
**CNN-based design.** ... (bold lead-in for sub-grouping)
**Transformer-based design.** ...
**Mamba and state-space design.** ...

### Inductive priors
**Topology-aware design.** ...
**Multi-task design.** ...
**Graph-based design.** ...

### Data regime
**Self-supervised pre-training.** ...
**Foundation models.** ...
**Federated learning.** ...
**Physics-informed models.** ...

(Table 2: representative methods with modality / family / dataset / metric)

## Downstream applications
### [Application 1]
### [Application 2]
### [Application 3]

## Translation to clinical practice
(Table 3: commercial products with regulatory + validation)

## Outstanding challenges

## Future directions

## References

Notes:

• No number prefixes on headings unless the journal requires them.
• In narrative AI method surveys, §Methods is usually 3 H3 subsections (the three axes), with bold lead-ins for each method family inside.
• In systematic/scoping reviews, use the structure in references/REVIEW_TYPES.md instead of forcing the 3-axis method taxonomy.
• Tables 1, 2, 3 are typically enough. Box 1 (metrics) is typical. Avoid 5+ tables.
• Verdict sentences cluster at the end of §Methods axis subsections and at the end of clinical translation discussions — not after every paragraph.

Verdict Sentences

For narrative reviews and method surveys, each major method-axis subsection (Architectural priors / Inductive priors / Data regime) should close with one verdict sentence expressing authorial position. Choose the 3-5 most opinionated positions across the whole manuscript — don't put verdicts on every paragraph.

For systematic and scoping reviews, verdicts must be constrained by the protocol and evidence map. Prefer "the included studies show..." over broad field-wide claims unless the search was designed to support the broader claim.

Verdict templates:

• "[Family] is currently the most cost-effective design choice for [problem]."
• "[Family] has yet to demonstrate clear advantage over [alternative] in clinical-grade evaluations."
• "[Family] is best understood as complementary to [alternative], not a replacement."
• "The next [N] years will determine whether [family] becomes the default backbone or remains a research curiosity."

Neutral catalogue is the LLM default and exactly what flagship review editors push back on. Force yourself to take 3-5 positions.

Required Elements

• Review type declaration before writing starts.
• Key Points box (4-5 bullets, 1-3 sentences each) after the title for narrative/flagship-style manuscripts.
• Tables 1-3 for narrative/method surveys: datasets, methods, commercial products.
• Systematic/scoping tables when applicable: search strategy, study characteristics, extraction variables, risk-of-bias summary.
• Box 1: evaluation metrics with formulas when useful; for systematic reviews, move formal methods definitions into Methods if the target journal prefers that.
• Figures: typically 3-5 for narrative reviews; systematic/scoping reviews require a PRISMA-style flow diagram.
• References: cite only what supports the argument. Quantity is downstream of substance — don't pad to a target count.
• Verdict sentences: 3-5 across narrative/method surveys, clustered at axis-section ends.
• Audit report: run the bundled scripts/audit_manuscript.py before delivery (resolve the path relative to this skill directory). The script is a triage tool — it flags likely issues from surface patterns; it does not prove any citation or number is correct. Delivery requires both a clean script pass (0 critical/high) and a manual source-level spot-check of quantitative and directional claims. A green script alone is not sufficient.

Formatting Quick Reference

Full rationale is in Core Principles above; this is the at-a-glance recap.

• Heading depth — max 2 body levels (H2/H3); no number prefixes unless journal-required; deeper grouping via bold lead-in **Topic.**; systematic/scoping Methods may follow PRISMA/journal conventions. (details)
• Equations — display equations ($$…$$) live in Box 1 (rarely additional Boxes); textbook formulas with no methodological insight go in prose, not inline. (details)
• Vendor names — Table 3 by default; sparse body mentions only where regulatory or comparative precision requires them. (details)

Citation Style

# Data citation
"...achieved Dice of 0.730 on ImageCAS [N]"

# Method citation
"Xu et al. [N] introduced..."

# Multi-citation (max 4 in one bracket — beyond that, regroup the claim)
"Multiple groups demonstrated this effect [N1, N2, N3]"

# Comparative
"While [N1] focused on architecture, [N2] addressed the data side"

[N] in body must match the bibliography entry [N], and bibliography [N] must be the paper the body sentence is actually attributing the claim to. See references/CITATION_INTEGRITY.md Rule 3.

Literature Sources

Use source types in combination. Confirm which tools are available in the current environment before using tool-specific names.

For closed-access journals (Med Image Anal, Eur Radiol, Lancet family) the user's local Zotero library is often the only path. Always check Zotero before assuming a paper is inaccessible.

For tool-adapter guidance, see references/MCP_SETUP.md.

Reference Files

Related Skills

For revising an existing AI-drafted review (whether your own previous output or someone else's draft), use ai-review-revision if it is installed. That skill is the dedicated tool for fixing draft-quality issues — multi-agent diagnostic, factual reset, structural reset, content polish, submission prep.

This skill (medical-imaging-review) is the dedicated tool for producing draft-quality content correctly the first time. They are complementary:

• medical-imaging-review = write-side (produce submission-quality first draft)
• ai-review-revision = revise-side (rescue a draft that already has quality issues)

If a draft produced by this skill still ends up needing the ai-review-revision workflow to land, that's a bug — flag it so this skill can be improved.

Version Notes

v3.0.0 was rewritten after the coronary-cta-paper draft exposed recurring failure modes: placeholder DOIs, citation drift, fabricated method modules, wrong performance numbers, vendor-style citations, flat method taxonomy, and AI-tone hedging.

v3.1.0 adds review-type routing, reporting-standard guidance, tool portability, softer structure rules, CCTA terminology correction, and an executable manuscript audit script.

v3.2.0 hardens the guardrails: the audit script now detects author↔citation mismatches under standard "Author et al. [N]" typesetting and recognises internationalised reference headings (## 参考文献, etc.) so Chinese drafts no longer mis-flag every citation; a fixture test suite (scripts/tests/) locks these in. Hard factual errors are now zero-tolerance (not gated behind a "5-or-more" threshold), unverified priority/novelty claims are forbidden, Phase 5 peer review is rewritten as executable sub-agent passes, and DOMAINS.md gains a generative/multimodal (VLM, diffusion, promptable-segmentation) paradigm section.

Consolidated fix ledger (v3.0.0 → v3.2.0):