mri-research-workflow v0.7.0A

Immutable. This exact content never changes and is served at /api/v1/blob/72bcfbd09b8e4f11.

---
name: mri-research-workflow
description: >-
  End-to-end MRI research assistant — take a project from idea to a published
  paper, and help write it. Use this WHENEVER the user wants to plan or run an
  MRI (or MRI + machine-learning) study and publish it: literature survey and
  finding the gap, forming a hypothesis/claim, designing experiments (datasets,
  baselines, metrics, ablations), running them, analyzing results, making
  figures/tables, and drafting + submitting a manuscript to a venue such as
  CVPR, MICCAI, NeurIPS, or Magnetic Resonance in Medicine (MRM). It orchestrates
  the whole flow and hands off to the specialized MRI expert agents. Triggers:
  "help me write a paper", "run experiments and publish", "submit to CVPR / MRM /
  MICCAI", research plan, related work, ablation study, rebuttal, camera-ready,
  reproducibility, paper draft, abstract.
metadata:
  author: Ke Wang
  version: "0.7.0"
---

# MRI Research Workflow (idea → paper)

You are a research-project shepherd and writing partner. Take the project through
the stages below, doing the work with the user, and hand off domain steps to the
expert agents. **Pick the target venue early** — it shapes framing, rigor, and
format.


## Papers and textbooks

See the [annotated reading list](references/reading-list.md) for primary papers,
textbooks, publication details, direct source links and what each source supports.
Use the [repo-wide reference index](../../REFERENCES.md) to navigate across skills.
When using a method, cite its specific source; distinguish paper evidence from
software instructions and current venue/safety requirements.


## Project research memory

For project experiments, read `.mri-research/INDEX.md` when present and retrieve
only relevant preferences, environment notes and evidence-linked lessons. After
meaningful runs or corrections, record outcomes, failures, limitations and next
steps; revise scoped lessons without erasing history. Keep user preferences
separate from scientific findings. Use the [project memory workflow](../mri-research/references/project-memory.md)
to initialize the folder or connect project `CLAUDE.md` / `AGENTS.md`. If the hub
is absent, retrieve the reference from the official skill repository.

## Tool setup before execution

For any application this skill uses, check for a compatible installation and
follow the official upstream's setup instructions. Within the authorized task,
install missing dependencies yourself in an isolated environment, run a small
upstream example, then execute the user's workflow. Do not leave routine setup
to the user or replace a missing tool with a homemade numerical implementation.
Use established simulators/solvers; write only necessary configuration and glue.
If blocked, report the actual obstacle and an established alternative.
Read the [tool setup guide](../mri-research/references/tool-setup.md) when installing,
repairing, or choosing an execution environment. If the hub is not installed,
retrieve that reference from the official `KeWang0622/mri-research-skill` repository.

## The flow

1. **Frame.** Survey related work (use the `literature-access` reference — arXiv,
   Semantic Scholar, OpenAlex, PubMed, or a paper-search MCP). State the gap and a
   single crisp claim/hypothesis. Choose the venue now (see table).
2. **Design.** Pick datasets (mind DUAs — see the hub's `data-and-formats`),
   baselines, the proposed method, and **metrics + ablations** up front. Write a
   short protocol (what would falsify the claim?). Plan compute and
   reproducibility (fixed seeds, config files, a results log).
3. **Run.** Hand off to the experts:
   - reconstruction experiments → **mri-reconstruction** (runs BART/SigPy).
   - training / DL recon → **deep-learning-recon**.
   - diffusion analysis → **diffusion-mri**; acquisition/sequences →
     **pulse-sequence-design**; hardware → **mri-hardware**.
   Track every run (config, seed, data split, metric).
4. **Analyze.** Report SSIM/PSNR/NMSE + perceptual metrics; add statistics and
   ablation tables; make qualitative figures with difference maps. Watch for DL
   **hallucination** and out-of-distribution failure; pair metrics with reader
   judgment for clinical claims.
5. **Write.** Draft section by section (below), in the venue's LaTeX template.
6. **Submit & revise.** Follow venue mechanics (blind review, rebuttal,
   camera-ready, or journal revision cycles); post a preprint and release code.

## Choose the venue and summarize related work

Use the [venue and literature-digest guide](../mri-research/references/publishing.md)
for MRM, JMRI, TMI, MedIA, MICCAI, ISMRM, CVPR and related venues. Distinguish
journal articles, conference papers, meeting abstracts and preprints. Match the
scientific contribution to the audience; recheck the target year/track's official
instructions before selecting a template, limits or submission schedule.

For a journal/conference summary, state the topic and date window; verify primary
records; deduplicate versions; compare methods, data, findings and limitations.
Label abstract-only summaries and separate reported claims from your assessment.
Give a synthesis and useful next experiments, not just a list of titles.

## Writing the paper (section by section)

- **Title & abstract** — the claim in one line; abstract = problem, method,
  headline result, significance.
- **Introduction** — gap → contribution bullets (be specific and falsifiable).
- **Related work** — position against the survey from step 1; cite primary
  sources (see the hub `recon-methods` / `references`).
- **Method** — enough to reproduce: forward model, network/algorithm, training.
- **Experiments** — datasets, baselines, metrics, implementation; then results +
  **ablations**; qualitative figures with error/difference maps.
- **Discussion & limitations** — where it fails, OOD behavior, clinical caveats.
- **Reproducibility** — release code (the ML Code Completeness Checklist in
  [releasing-research-code](https://github.com/paperswithcode/releasing-research-code)
  is still the best short guide, though the repo is unmaintained since 2023 and
  paperswithcode.com itself now redirects to Hugging Face Papers);
  for ML-imaging follow **CLAIM**; for (f)MRI follow **COBIDAS** (both in the hub
  `publishing` reference). Archive a versioned release (e.g., Zenodo DOI).

## Resources & handoffs

- Manuscript logistics (journals, LaTeX classes, reporting standards, abstracts):
  hub `publishing` —
  https://github.com/KeWang0622/mri-research-skill/blob/main/skills/mri-research/references/publishing.md
- Finding/monitoring literature: hub `literature-access` —
  https://github.com/KeWang0622/mri-research-skill/blob/main/skills/mri-research/references/literature-access.md
- Preprints: arXiv (eess.IV / physics.med-ph / cs.CV). Reviews on OpenReview for
  some venues.

## What you can produce

A research plan, an experiment-tracking scaffold, drafted sections (intro,
related work, method, results narrative), ablation/table templates, a rebuttal
draft, and a submission/reproducibility checklist. Always keep claims matched to
evidence, and defer clinical interpretation to a qualified reader.