henhaobio
Sign in

PLoS Computational BiologyNeurosciencePeer-reviewed journal article

Mindboggling morphometry of human brains

Arno KleiniD, Satrajit GhoshiD, Forrest Sheng BaoiD, Joachim Giard, Yrjö HämeiD, Eliezer Stavsky, Noah Lee, B Rossa, Martin ReuteriD, Elias Chaibub NetoiD, Anisha KeshavaniD

PDFDOIPreprintCodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

Mindboggle (http://mindboggle.info) is an open source brain morphometry platform that takes in preprocessed T1-weighted MRI data and outputs volume, surface, and tabular data containing label, feature, and shape information for further analysis. In this article, we document the software and demonstrate its use in studies of shape variation in healthy and diseased humans. The number of different shape measures and the size of the populations make this the largest and most detailed shape analysis of human brains every conducted. Brain image morphometry shows great potential for providing much-needed biological markers for diagnosing, tracking, and predicting progression of mental health disorders. Very few software algorithms provide more than measures of volume and cortical thickness, and more subtle shape measures may provide more sensitive and specific biomarkers. Mindboggle computes a variety of (primarily surface-based) shapes: area, volume, thickness, curvature, depth, Laplace-Beltrami spectra, Zernike moments, etc. We evaluate Mindboggles algorithms using the largest set of manually labeled, publicly available brain images in the world and compare them against state-of-the-art algorithms where they exist. All data, code, and results of these evaluations are publicly available.\n\nAuthor SummaryBrains vary in many ways, including their shape. Analysing differences in shape between brains or changes in brain shape over time has been used to characterize morphology of diseased brains, but these analyses conventionally rely on simple volumetric shape measures. We believe that access to a greater variety of shape measures could provide greater sensitivity and specificity to morphological disturbances, and could aid in diagnosis, tracking, and prediction of the progression of mental health disorders. Mindboggle is open source software that provides neuroscientists (and indeed, anyone interested in computing shapes) tools for computing a variety of shape measures, including area, volume, thickness, curvature, geodesic depth, travel depth, Laplace-Beltrami spectra, and Zernike moments. In addition to algorithmic contributions, we conducted evaluations and applied Mindboggle to conduct the most detailed shape analysis of human brains.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (3 Dec 2016) · CC BY 4.0

Open source Mindboggle pipeline extracts cortical folds, sulci and fundi and computes diverse shape measures, yielding normative morphometry for 101 manually labeled healthy human brains.

Brain morphometry usually relies on a few measures such as volume, area and cortical thickness, and on registration that does not guarantee anatomical correspondence, which limits biomarker sensitivity. Mindboggle adds surface-based depth, curvature, spectral and moment descriptors on labeled gyri, sulci and fundi, and evaluates them against FreeSurfer, ANTs and alternative fundus extractors.

Key results

  1. Shownthickinthehead, a volume-based cortical thickness estimate that avoids surface reconstruction, agreed better with published manual MRI thickness ranges than FreeSurfer or ANTs did.

    In 40 EMBARC control subjects over 16 cortical regions, almost ninety percent of thickinthehead estimates fell inside published ranges from manual MRI delineation, versus forty percent for FreeSurfer; ANTs deviated further and had larger inter-scan and inter-subject ranges. · 3.1.2. Comparison between cortical thickness…; 4.1 Summary of results

  2. ShownRelated surface measures are not independent and should stand in for one another only with care: geodesic depth inflates insular depth relative to travel depth, and travel depth represents shallow cortex more faithfully than FreeSurfer convexity.

    Travel vs geodesic depth distance correlation 0.91 to 1.00 in 60 of 62 regions, but 0.29 and 0.31 in left and right insula (101 subjects); mean vs FreeSurfer curvature 0.73 (insula) to 0.91 (rostral middle frontal). On one surface, 97.71% of small-convexity vertices were near the wrapper yet made up only 6.89% of near-wrapper vertices. · 3.1. Comparisons between brain shape measures; 3.1.1. Comparison between travel depth and…; Tables

  3. ShownIn the first quantitative comparison of sulcal fundus extraction, Mindboggle fundi were about as close to manually edited DKT label boundaries as Gang Li's. Forrest Bao's tree-based fundi were closer in the reverse direction, so no method dominated.

    Over 48 DKT sulci in Mindboggle-101, boundary-to-fundus mean distance: Gang Li 2.09mm, Mindboggle 2.38mm, Forrest Bao 3.65mm (closest in 25, 21 and 2 sulci); fundus-to-boundary: Bao 3.33mm vs Mindboggle 4.06mm and Li 4.65mm (41 vs 5 and 2 sulci). · 3.2. Evaluation of fundus extraction algorithms

  4. ShownMost shape measures were reproducible across repeat scans of the same person, and differences between left and right hemispheres exceeded differences between scans.

    41 rescanned subjects (OASIS-TRT-20, MMRR-21), 31 left regions: among volumetric measures only entorhinal volume reached a fractional difference of at least 0.10 (0.21); largest were FreeSurfer curvature in pericalcarine (0.34), insula (0.28) and rostral anterior cingulate (0.23). Interhemispheric differences across 101 brains were much higher. · 3.3. Consistency of shape measures between MRI…; 3.4. Measuring shape differences between left…; Tables

  5. SuggestedAcross 101 healthy brains from five laboratories, most shape variability lay within hemispheres (the residual) rather than between hemispheres, subjects or labs. In ADNI, changes in several regional measures over time tracked changes in cognitive score.

    24 linear mixed models per measure-region pair (fixed: sex, handedness, age; random: lab, subject, hemisphere) ranked by BIC; for four measures (not geodesic depth) the two best models kept handedness and age. ADNI (126 Alzheimer's, 199 controls): three-year changes in volume, FreeSurfer thickness and mean curvature (e.g., entorhinal, temporal, precuneus) correlated with ADNI-MEM change. · 3.7. Measuring human brain shape variation; 4.1 Summary of results

Methods, data and reproducibility

A Python 3/C++ Nipype pipeline ingests FreeSurfer (optionally ANTs) outputs from T1-weighted MRI and can merge their gray/white segmentations. It computes regional volume and thickinthehead thickness plus per-vertex area, curvature, geodesic and travel depth, then extracts folds, sulci and fundi and derives per-region Laplace-Beltrami spectra and Zernike moments.

  • Mindboggle-101: T1-weighted MRI with manually edited DKT cortical labels, used for measure comparisons, fundus evaluation and normative shape analysis · human · n = 101 individuals (20 with CMA non-cortical labels)
  • Repeat-scan subset of Mindboggle-101 drawn from the OASIS-TRT-20 and MMRR-21 groups, used to assess scan-rescan shape consistency · human · n = 41 subjects
  • EMBARC control subjects for cortical thickness reliability and accuracy; further EMBARC images visually inspected for FreeSurfer segmentation errors · human · n = 40 control subjects; over 100 images inspected · NCT01407094
  • ADNI participants scanned twice three years apart, used to relate shape change to ADNI-MEM cognitive score change · human · n = 126 with Alzheimer's, 199 healthy controls
  • AddNeuroMed data processed alongside ADNI for an international Alzheimer's disease challenge · human
Methods & tools
Mindboggle (Python 3, Nipype, Docker container), FreeSurfer v5.1 recon-all (version 6 recommended), ANTs antsCorticalThickness.sh, Atropos, ImageMath (v2.1.0rc3 or higher), Nipype, VTK (Visualization Toolkit), Desikan-Killiany-Tourville (DKT) labeling protocol; DKT-40/DKT-100 atlases, Shape-DNA Laplace-Beltrami spectra (Reuter); Weighted Spectral Distance (WESD), 3-D Zernike moments (Pozo et al. implementation), BrainVisa, Forrest Bao's pruned minimum spanning tree fundus algorithm, Gang Li's fundus extraction algorithm, Linear mixed models with Bayesian Information Criterion selection
Software & environment
Python 3 with Python-wrapped C++ in Nipype; tested most with Python 3.5.1 on Ubuntu Linux 14.04; Docker container available; inputs from FreeSurfer v5.1 (version 6 recommended); ANTs v2.1.0rc3 or higher recommended.
Experimental conditions
Travel depth on a mesh of 150,000 vertices and 300,000 triangles takes around 200 seconds on an ordinary computer once the wrapper surface exists; building the wrapper adds 20 seconds at a 5 mm probe radius.
Reproducibility gaps
Native install targets Linux (a cross-platform Docker container exists) and was tested mainly on one Python/Ubuntu setup; access conditions for ADNI, AddNeuroMed and EMBARC images are not described; many detailed results are deferred to OSF pages and supplements.
Limitations (6)
  • The optional FreeSurfer/ANTs hybrid segmentation still over-includes non-brain tissue and can replace true gray matter with white matter where surface reconstruction fails. · stated by the authors
  • Fundus evaluation lacks ground truth: manual DKT label boundaries served as a proxy, BrainVisa was excluded for extracting too few fundi, no method clearly won, and the diagnostic utility of fundi was left untested. · stated by the authors
  • Scan-rescan consistency was weaker for small structures (entorhinal volume difference 0.21) and for FreeSurfer curvature, whose sign changes can inflate fractional differences where median curvature is near zero. · stated by the authors
  • Feature extraction rests on hand-designed structure and label definitions whose validity is assumed; atypical pathological brains (tumors, lesions, edema) may fit poorly, which motivates learned or unsupervised features. · stated by the authors
  • The text reports ADNI associations with ADNI-MEM change as significant but gives no effect sizes, p-values or multiple-comparison correction, and the associations rest on averaged ranks of different tests, so their robustness cannot be judged here. · noted in this summary
  • Thickness accuracy was judged against published manual ranges for only 16 regions in 40 controls, not against manual or histological measurements on the same brains, so agreement could reflect population differences. · noted in this summary

Read this if you need surface-based cortical shape measures (travel depth, sulcal fundi, Laplace-Beltrami spectra, Zernike moments) beyond FreeSurfer volume and thickness, or normative shape data for 101 labeled brains.

Citations
723
FWCI
12.71
Field percentile
top 1%

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISIm '25Potvin '21Manera '20Germann '25Im '26Im '25
63% semantic matchNeuroscience
Yanghee Im, Leila Nabulsi, Melody J.Y. Kang+70 more·Lecture notes in computer science·Oct 4, 2025