Phenotype discovery of traumatic brain injury segmentations from heterogeneous multi-site data

Adam M. Saunders; Bennett A. Landman; Chloe Cho; Daniel R. Gillis; Elyssa M. McMaster; Gaurav Rudravaram; Lianrui Zuo; Lucas W. Remedios; Michael E. Kim; Tonia S. Rex

arxiv: 2511.03767 · v1 · submitted 2025-11-05 · 🧬 q-bio.QM · eess.IV

Phenotype discovery of traumatic brain injury segmentations from heterogeneous multi-site data

Adam M. Saunders , Michael E. Kim , Gaurav Rudravaram , Lucas W. Remedios , Chloe Cho , Elyssa M. McMaster , Daniel R. Gillis , Yihao Liu

show 3 more authors

Lianrui Zuo Bennett A. Landman Tonia S. Rex

This is my paper

Pith reviewed 2026-05-18 01:33 UTC · model grok-4.3

classification 🧬 q-bio.QM eess.IV

keywords traumatic brain injuryMRImulti-site databrain segmentationregion of interestindependent component analysisphenotype discoveryvolume analysis

0 comments

The pith

Harmonized multi-site MRI data reveals volume differences in 37 brain regions between TBI patients and controls.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper processes thousands of T1-weighted MRI scans from the FITBIR repository across multiple sites to uncover shared structural patterns in traumatic brain injury. It first harmonizes the images to reduce site-specific artifacts, segments 132 regions of interest, computes z-scores of volumes relative to controls, and adjusts for age, sex, and total brain volume. Statistical tests then identify significant differences in 37 regions, which independent component analysis groups into three clusters: posterior structures including the brainstem and occipital pole, subcortical gray matter and insular cortex, and cerebral plus cerebellar white matter. A reader would care because TBI outcomes vary greatly and linking damage locations to functional effects could support more precise diagnosis and tracking.

Core claim

By harmonizing heterogeneous multi-site T1-weighted images to a local dataset and segmenting 132 regions of interest, the study calculates volumes and z-scores for all participants. After regressing out sex, age, and total brain volume, independent t-tests with false discovery rate correction find significant differences in 37 ROIs between TBI subjects and controls. Independent component analysis applied to the component loadings of TBI cases shows these differences arise from three main patterns: the brainstem, occipital pole and structures posterior to the orbit; subcortical gray matter and insular cortex; and cerebral and cerebellar white matter.

What carries the argument

Harmonization of multi-site T1-weighted MRI followed by automated segmentation of 132 ROIs, z-score normalization against controls, and independent component analysis on the resulting volume data to extract injury phenotypes.

If this is right

The three identified injury patterns could serve as candidate phenotypes for stratifying TBI patients in future clinical trials.
Volume changes in these ROIs might be tracked longitudinally to monitor recovery or progression after injury.
The pipeline supports combining data from new sites without requiring identical scanners or protocols.
Correlating the component loadings with clinical scores could strengthen links between structure and function.
Outlier removal and quality assurance steps reduce the chance that site-specific noise drives the reported differences.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Extending the approach to include diffusion or functional MRI could test whether the same regions show connectivity changes.
Applying the method to pediatric or mild TBI cohorts might reveal age-specific patterns not captured here.
If the phenotypes prove stable, they could be used to select patients for targeted rehabilitation focused on affected pathways.
Testing the pipeline on other heterogeneous neurological datasets would show whether the harmonization plus ICA step generalizes beyond TBI.

Load-bearing premise

Harmonization of heterogeneous multi-site T1-weighted images combined with automated segmentation of 132 ROIs produces volume measurements that accurately reflect true structural differences rather than acquisition artifacts or segmentation errors.

What would settle it

Re-running the group comparisons on the same scans without the harmonization step or against a small set of manually segmented volumes to check whether the 37 significant ROIs remain or vanish.

Figures

Figures reproduced from arXiv: 2511.03767 by Adam M. Saunders, Bennett A. Landman, Chloe Cho, Daniel R. Gillis, Elyssa M. McMaster, Gaurav Rudravaram, Lianrui Zuo, Lucas W. Remedios, Michael E. Kim, Tonia S. Rex, Yihao Liu.

**Figure 1.** Figure 1: The FITBIR repository contains large-scale multi-site data from subjects with TBI and controls from a large range of ages. Across 25 studies, there are 8,212 MRI sessions. 1,882 sessions are from controls, 5,811 are from those with TBI, and 519 have an unknown TBI status. Only sessions where age is available are shown [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 2.** Figure 2: Images from three example studies in the FITBIR repository demonstrate the heterogeneous presentation in structural MRI data. The FITBIR repository contains imaging data from studies with highly variable acquisition parameters and contrast. Some contrasts are missing, such as the T2-weighted image from Study 226. The resolutions vary across the through-plane and in-plane of the images. We show sagittal and… view at source ↗

**Figure 3.** Figure 3: Using HACA3, we harmonized data from each session to a T1 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 4.** Figure 4: After applying our preprocessing pipeline using super-resolution and image harmonization, the same three T1-weighted images shown in [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

**Figure 5.** Figure 5: We identified 37 ROIs with significant differences in volume in TBI compared to controls (independent t [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗

**Figure 6.** Figure 6: The 37 ROIs with significant differences involve the white matter, cerebellum, brainstem, and structures [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗

**Figure 7.** Figure 7: We ran independent component analysis on the z [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗

read the original abstract

Traumatic brain injury (TBI) is intrinsically heterogeneous, and typical clinical outcome measures like the Glasgow Coma Scale complicate this diversity. The large variability in severity and patient outcomes render it difficult to link structural damage to functional deficits. The Federal Interagency Traumatic Brain Injury Research (FITBIR) repository contains large-scale multi-site magnetic resonance imaging data of varying resolutions and acquisition parameters (25 shared studies with 7,693 sessions that have age, sex and TBI status defined - 5,811 TBI and 1,882 controls). To reveal shared pathways of injury of TBI through imaging, we analyzed T1-weighted images from these sessions by first harmonizing to a local dataset and segmenting 132 regions of interest (ROIs) in the brain. After running quality assurance, calculating the volumes of the ROIs, and removing outliers, we calculated the z-scores of volumes for all participants relative to the mean and standard deviation of the controls. We regressed out sex, age, and total brain volume with a multivariate linear regression, and we found significant differences in 37 ROIs between subjects with TBI and controls (p < 0.05 with independent t-tests with false discovery rate correction). We found that differences originated in 1) the brainstem, occipital pole and structures posterior to the orbit, 2) subcortical gray matter and insular cortex, and 3) cerebral and cerebellar white matter using independent component analysis and clustering the component loadings of those with TBI.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This paper runs standard harmonization, ROI volumetry, regression, and ICA on a large FITBIR multi-site TBI cohort to report 37 differing regions and three clusters, but provides no validation metrics for the pipeline.

read the letter

This paper pulls T1 images from 25 studies in the FITBIR repository, harmonizes them to a local set, segments 132 ROIs, z-scores volumes against controls, regresses out age, sex, and total brain volume, then flags 37 regions with FDR-corrected t-test differences. ICA on the TBI cases plus clustering of loadings points to three patterns: brainstem and structures behind the orbit, subcortical gray and insula, and cerebral plus cerebellar white matter. The scale of the public dataset is the clear strength here, and the steps follow established practice for handling site variation and multiple testing. The concrete regional counts and cluster descriptions give a data-driven view of injury patterns that prior smaller studies have only sketched. The soft spot sits in the unverified pipeline. TBI brains often contain lesions or atrophy that degrade automated segmentation, yet the work reports no Dice scores, manual label comparisons, or post-harmonization site-effect plots. Harmonizing everything to one local reference could leave residual batch effects that mimic the reported volume shifts. Outlier rules and the choice of ICA components also stay vague in the abstract. This is useful for groups already running similar TBI imaging pipelines who want to see what emerges from a big heterogeneous sample. It does not introduce new methods but supplies specific numbers from this cohort that could inform stratification ideas. I would send it for peer review so referees can check the full methods, any validation numbers, and code. The observational claims are straightforward enough that detailed review can decide if the findings hold once the measurement steps are confirmed.

Referee Report

1 major / 2 minor

Summary. The manuscript analyzes heterogeneous multi-site T1-weighted MRI data from the FITBIR repository (25 studies, 7,693 sessions) to discover phenotypes of traumatic brain injury (TBI). The pipeline involves harmonizing images to a local dataset, segmenting 132 regions of interest (ROIs), quality assurance, outlier removal, z-scoring volumes relative to controls, multivariate regression for covariates (sex, age, total brain volume), and identifying 37 significant ROI differences using independent t-tests with FDR correction. Independent component analysis (ICA) followed by clustering of component loadings in TBI subjects reveals three clusters of injury origins: (1) brainstem, occipital pole and structures posterior to the orbit; (2) subcortical gray matter and insular cortex; (3) cerebral and cerebellar white matter.

Significance. If the harmonization and segmentation steps are shown to be valid, the work could contribute to phenotype discovery in heterogeneous TBI by exploiting a large public dataset and standard tools (multivariate regression, FDR-corrected t-tests, ICA, and clustering). Strengths include use of external public data, large sample size, and absence of circular derivations or self-referential parameters. The reported regional patterns would then provide falsifiable structural hypotheses for follow-up studies.

major comments (1)

[Methods (harmonization and segmentation)] Methods (harmonization and segmentation): The pipeline harmonizes multi-site T1 images (varying resolutions/parameters across 25 studies) to an unspecified local dataset then applies automated segmentation of 132 ROIs, yet reports no validation metrics (e.g., Dice scores, manual-label correlations, or post-harmonization site-effect diagnostics such as residual batch-effect tests). TBI pathology (lesions, atrophy) is known to degrade standard automated tools; without these checks the 37 significant ROI differences and the three ICA-derived clusters cannot be confidently attributed to biology rather than acquisition artifacts or segmentation failure.

minor comments (2)

[Abstract] Abstract: The number of ICA components and the clustering method (e.g., k-means, hierarchical) are not stated; it is unclear whether component count was pre-specified or data-driven.
[Abstract] Abstract: Outlier removal and quality-assurance criteria are mentioned but not specified (e.g., volume thresholds, visual inspection protocol), limiting reproducibility.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive comments on our manuscript. We address the major concern regarding validation of the harmonization and segmentation pipeline below.

read point-by-point responses

Referee: Methods (harmonization and segmentation): The pipeline harmonizes multi-site T1 images (varying resolutions/parameters across 25 studies) to an unspecified local dataset then applies automated segmentation of 132 ROIs, yet reports no validation metrics (e.g., Dice scores, manual-label correlations, or post-harmonization site-effect diagnostics such as residual batch-effect tests). TBI pathology (lesions, atrophy) is known to degrade standard automated tools; without these checks the 37 significant ROI differences and the three ICA-derived clusters cannot be confidently attributed to biology rather than acquisition artifacts or segmentation failure.

Authors: We appreciate the referee highlighting the importance of validation for harmonization and segmentation. The manuscript states that quality assurance and outlier removal were performed after segmentation of the 132 ROIs. We agree that more explicit metrics and diagnostics would strengthen the work. In the revised manuscript we will specify the harmonization procedure and reference dataset, add post-harmonization site-effect diagnostics (e.g., residual batch-effect tests or PCA on volumes), and expand the discussion to address potential effects of TBI pathology on automated segmentation while describing how our QA and outlier removal steps were designed to mitigate segmentation failures. We will also cite published validation performance of the segmentation tool in TBI cohorts. These additions will support attribution of the 37 ROI differences and three clusters to biological patterns. revision: partial

Circularity Check

0 steps flagged

No circularity: standard empirical pipeline on external data

full rationale

The paper describes an empirical analysis pipeline on the public FITBIR repository: harmonization of multi-site T1 images to a local dataset, automated segmentation of 132 ROIs, outlier removal, z-scoring against controls, covariate regression, FDR-corrected t-tests, ICA, and clustering of component loadings. No mathematical derivations, first-principles predictions, or fitted parameters are presented that reduce by construction to the inputs or to self-citations. All steps rely on external public data and standard statistical tools with no load-bearing self-citation chains, uniqueness theorems, or ansatzes that define the reported ROI differences or clusters. The central claims remain independent of any circular reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central findings rest on standard statistical and neuroimaging assumptions rather than new free parameters or invented entities. No numbers are fitted beyond conventional regression and FDR thresholds.

axioms (2)

domain assumption Controls provide an appropriate reference distribution for computing z-scores of ROI volumes in TBI subjects.
Z-scores are explicitly calculated relative to the mean and standard deviation of the control group.
domain assumption Multivariate linear regression fully removes confounding effects of sex, age, and total brain volume on ROI volumes.
Regression is applied prior to group comparisons and ICA.

pith-pipeline@v0.9.0 · 5851 in / 1449 out tokens · 39146 ms · 2026-05-18T01:33:06.202703+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We regressed out sex, age, and total brain volume with a multivariate linear regression, and we found significant differences in 37 ROIs between subjects with TBI and controls (p < 0.05 with independent t-tests with false discovery rate correction). We found that differences originated in 1) the brainstem, occipital pole and structures posterior to the orbit, 2) subcortical gray matter and insular cortex, and 3) cerebral and cerebellar white matter using independent component analysis and clustering the component loadings of those with TBI.

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

2 extracted references · 2 canonical work pages

[1]

cone of vulnerability

RESULTS After applying our preprocessing involving super-resolution and image harmonization, the resulting set of synthesized T1-weighted images was more consistent in anatomical contrast and resolution (Fig. 4). The gray matter-white matter contrast was enhanced, despite the poor contrast available in the T1-weighted images for Studies 226 and 246. The U...

work page 2021
[2]

N4ITK: Improved N3 Bias Correction,

Tustison, N. J., Avants, B. B., Cook, P. A., Yuanjie Zheng, Egan, A., Yushkevich, P. A. and Gee, J. C., “N4ITK: Improved N3 Bias Correction,” IEEE Trans Med Imaging 29(6), 1310–1320 (2010). [26] Fonov, V., Evans, A. C., Botteron, K., Almli, C. R., McKinstry, R. C. and Collins, D. L., “Unbiased average age-appropriate atlases for pediatric studies,” Neuroi...

work page 2010

[1] [1]

cone of vulnerability

RESULTS After applying our preprocessing involving super-resolution and image harmonization, the resulting set of synthesized T1-weighted images was more consistent in anatomical contrast and resolution (Fig. 4). The gray matter-white matter contrast was enhanced, despite the poor contrast available in the T1-weighted images for Studies 226 and 246. The U...

work page 2021

[2] [2]

N4ITK: Improved N3 Bias Correction,

Tustison, N. J., Avants, B. B., Cook, P. A., Yuanjie Zheng, Egan, A., Yushkevich, P. A. and Gee, J. C., “N4ITK: Improved N3 Bias Correction,” IEEE Trans Med Imaging 29(6), 1310–1320 (2010). [26] Fonov, V., Evans, A. C., Botteron, K., Almli, C. R., McKinstry, R. C. and Collins, D. L., “Unbiased average age-appropriate atlases for pediatric studies,” Neuroi...

work page 2010