The Methods Companion : Lesion Network Mapping

How this works Eight Decisions · One Synthesis

A scaffolded conversation
for designing your study.

A structured walk through the core methodological choices that shape a lesion network mapping paper, from phenomenon to validation, with synthesis at the end.

For researchers, trainees, and methodologists Decisions are saved in your browser session

What this is. A planning tool. You answer eight questions about the study you are scoping. The companion then drafts a methods skeleton tailored to your choices, flags decisions worth reconsidering, and stays available for follow-up questions.

What it is not. A substitute for your own judgment, your collaborators, or the actual literature. Treat the synthesis as a first draft to argue with.

i. Phenomenon

What behavior, symptom, or condition are you mapping?

Name it as specifically as you can. Mystical experience, post-stroke depression, expressive aphasia, religious fundamentalism, central hypersomnia. The sharper the construct, the cleaner the network.

ii. Outcome Structure

How is the outcome variable scaled?

This determines whether you are running a case versus control contrast, a continuous regression on lesion network maps, or something more bespoke.

A Binary or categorical Cases versus controls, presence versus absence, syndrome versus no syndrome. B Ordinal or ranked severity Symptom present in graded categories: mild, moderate, severe. C Continuous score on a validated scale PHQ-9, BDI, TAS-20, MEQ-30, ESS, or comparable continuous measures. D Multiple outcomes at once A primary outcome plus secondary measures, or several scales tapping the construct. E Or something else Describe the structure of your outcome variable.

iii. Lesion Source

Where do your lesions come from?

The provenance shapes everything downstream: tracing fidelity, sample heterogeneity, demographic coverage, and the kinds of claims you can defensibly make.

A Prospective in-house cohort You or your collaborators are recruiting and imaging patients directly. B Retrospective chart review Existing clinical scans plus chart documentation of the symptom or condition. C Published case reports and case series Literature mining with figure tracing or atlas-based localization from text descriptions. D Existing lesion database Vanderbilt, BWH, penetrating head injury cohorts, or another curated collection. E Combined or multi-source A discovery cohort from one source plus a replication cohort from another. F Or something else Describe your source briefly.

iv. Lesion Tracing

How will lesions be segmented?

Tracing strategy is one of the most underreported determinants of LNM reproducibility. Be explicit, and be prepared to defend it.

A Manual tracing on T1, T2, or FLAIR A trained rater segments each lesion in native space, then warps to MNI. B Manual with double-rater consensus Two raters trace independently, blinded to symptom status, with reconciliation of disagreements. C Atlas-based from published descriptions For literature-derived cases without scans: localizing reported sites onto an atlas. D Semi-automated segmentation Tools like LINDA, BIANCA, or LST, with manual quality control. E Pre-segmented by collaborators You inherit traced masks from the originating site. F Or something else Describe your tracing approach.

v. Connectome

Which connectome will you use as the reference?

The connectome is the substrate against which each lesion is queried. Functional, structural, normative, or disease-matched: each choice carries a different inferential commitment.

A Normative functional connectome Resting-state fMRI from a healthy reference cohort: GSP1000, HCP, Yeo, or similar. B Normative structural connectome Diffusion-derived, often DSI Studio HCP-842 or a comparable group-averaged template. C Both functional and structural A multimodal approach with parallel analyses and convergence checks. D Disease-matched connectome Reference connectome derived from a population matched to your patient demographics or pathology. E Or something else Describe your connectome choice.

vi. Statistical Approach

How will you derive the network?

This follows from your outcome structure. The companion will sanity-check the alignment between the two.

A Voxel-wise t-test, cases versus controls Two-sample contrast on lesion network maps with thresholding. B Voxel-wise regression on a continuous score Each voxel's connectivity is regressed against the symptom severity score across patients. C Permutation-based with FWE correction Non-parametric thresholding, robust to non-normal distributions, more conservative inference. D Mixed effects or covariate-adjusted Modeling lesion volume, age, time since lesion, or site as covariates. E Or something else Describe your statistical approach.

vii. Validation

How will you validate the network?

A network is only as strong as its replication. Pick one validation lane, or several.

☐ Leave-one-out cross-validation Iteratively withhold one patient, regenerate the network, predict the held-out symptom score. ☐ Split-half replication Random partition of the cohort, derive the network in one half, test in the other. ☐ Independent external cohort Replicate the network in a wholly separate dataset from a different site. ☐ Causal validation: DBS or TMS data Test whether stimulation of the network alters the symptom in independent neuromodulation cohorts. ☐ Or something else Describe your validation strategy. You can select more than one option above.

viii. Specificity

How will you establish specificity?

Showing your network exists is one thing. Showing it picks out your phenomenon, and not a generic lesion-network signature, is another.

☐ Comparison to a symptom library Spatial similarity against published lesion networks for other conditions. ☐ Negative control conditions A matched cohort with lesions but without the symptom of interest. ☐ Spatial correlation with published networks Quantify convergence or divergence with prior LNM findings in adjacent constructs. ☐ Construct-specific behavioral validation Show the network does not predict an unrelated behavior measured in the same cohort. ☐ Or something else Describe your specificity strategy. Select multiple if applicable.

Eight steps. The synthesis lives below.

Synthesis Tailored to your responses

A working draft of your methods.

Follow-up Continue the conversation

Lesion Network Mapping
AI Assistant

A scaffolded conversation
for designing your study.

What behavior, symptom, or condition are you mapping?

How is the outcome variable scaled?

Where do your lesions come from?

How will lesions be segmented?

Which connectome will you use as the reference?

How will you derive the network?

How will you validate the network?

How will you establish specificity?

A working draft of your methods.

Anything you want to push on?

Lesion Network MappingAI Assistant

What behavior, symptom, or condition are you mapping?

How is the outcome variable scaled?

Where do your lesions come from?

How will lesions be segmented?

Which connectome will you use as the reference?

How will you derive the network?

How will you validate the network?

How will you establish specificity?

A working draft of your methods.

Anything you want to push on?

Lesion Network Mapping
AI Assistant