pith. sign in

arxiv: 1907.05395 · v1 · pith:ZNCJ7C3Mnew · submitted 2019-07-11 · 🧬 q-bio.NC · eess.IV· q-bio.QM

Cortical Surface Parcellation using Spherical Convolutional Neural Networks

Prasanna Parvathaneni , Shunxing Bao , Vishwesh Nath , Neil D. Woodward , Daniel O. Claassen , Carissa J. Cascio , David H. Zald , Yuankai Huo
show 2 more authors
Bennett A. Landman Ilwoo Lyu
This is my paper

Pith reviewed 2026-05-24 22:37 UTC · model grok-4.3

classification 🧬 q-bio.NC eess.IVq-bio.QM
keywords cortical parcellationspherical convolutional neural networksdeformation fieldssurface registrationbrain mappingdeep learningneuroimaging
0
0 comments X

The pith

Spherical convolutional neural networks trained on deformation-augmented data achieve accurate cortical parcellation in under a minute.

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

The paper proposes using spherical deep convolutional neural networks for dividing the cortical surface into 49 labels. Traditional multi-atlas methods depend on inter-subject surface registration that takes hours and often fails to match parcel boundaries exactly to geometric features. The approach generates extra training examples by aligning ground-truth parcel boundaries, deriving deformation fields, and applying those fields to create new pairs of deformed geometric features with their corresponding parcellation maps, then smoothly morphing with intermediate fields. On a set of 427 adult brains this yields higher accuracy than both multi-atlas registration and naive spherical U-Net baselines while finishing the full parcellation in less than a minute. A sympathetic reader would care because it replaces a slow, registration-heavy pipeline with a direct learned mapping that could scale to large cohorts.

Core claim

The central claim is that cortical surface parcellation using spherical deep convolutional neural networks becomes feasible and superior when training data are expanded by aligning ground-truth parcel boundaries to produce deformation fields, generating new pairs of deformed geometric features and parcellation maps, and then smoothly morphing those maps with intermediate fields. This training regimen allows the networks to outperform traditional multi-atlas registration and naive spherical U-Net approaches on 427 adult brains for 49 labels while completing full parcellation in less than a minute.

What carries the argument

Spherical convolutional neural networks trained on pairs of geometric features and parcellation maps that have been augmented and morphed using deformation fields derived from ground-truth parcel boundary alignments.

If this is right

  • Full cortical parcellation into 49 labels is possible in less than one minute per subject.
  • The method outperforms both traditional multi-atlas registration (2-3 hours) and naive spherical U-Net baselines.
  • Training data augmentation via parcel-boundary deformation fields improves accuracy on a cohort of 427 adult brains.

Where Pith is reading between the lines

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

  • The same augmentation technique could be applied to other surface-based neuroimaging tasks that currently rely on slow registration.
  • If the learned mapping holds, large-scale population studies could obtain parcel labels without per-subject registration pipelines.
  • The results suggest that parcel boundaries carry information beyond the geometric features used in classical registration.

Load-bearing premise

The deformation fields derived from aligning ground-truth parcel boundaries produce training examples that generalize without bias to unseen subjects' geometric features.

What would settle it

A held-out test set of cortical surfaces whose geometric features lie outside the range of deformations used in training, showing measurably lower parcellation accuracy than the reported results, would falsify the generalization claim.

Figures

Figures reproduced from arXiv: 1907.05395 by Bennett A. Landman, Carissa J. Cascio, Daniel O. Claassen, David H. Zald, Ilwoo Lyu, Neil D. Woodward, Prasanna Parvathaneni, Shunxing Bao, Vishwesh Nath, Yuankai Huo.

Figure 1
Figure 1. Figure 1: An overview of the proposed method. Three geometric features (iH, SD, H) are used for training the spherical U-Net to predict 49 cortical parcellation labels. For each geometric property, intermediate deformation fields draw a total of 11+11 respective samples after boundary and geometric alignment for data augmentation. The cortical parcellation is then performed using the original geometric features of t… view at source ↗
Figure 2
Figure 2. Figure 2: Boundary extraction and alignment. (1st row) For inputs for training, parcel boundaries are obtained from ground-truth labels (Eq. (2)). The boundaries are used to generate distance map T by solving an eikonal equation, and (2nd row) smooth trajectory of its deformation to a template is represented by increasing spherical har￾monics degree l. (3rd row) The features for training are accordingly deformed by … view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparison: ground-truth, multi-atlas, spherical U-Net, and spher￾ical U-Net with the proposed features. Our approach shows better performance than the other methods. The arrows highlight the mismatching regions to the ground-truth. ACgG AIns AOrG AnG Calc CO Cun Ent FO FRP FuG GRe IOG ITG LiG LOrGMCgGMFCMFGMOGMOrGMPoGMPrGMSFGMTG 60 70 80 90 100 Dice (%) OCPOFuG OpIFG OrIFG PCgG PCu PHG PIns PO… view at source ↗
Figure 4
Figure 4. Figure 4: Dice overlap of 49 regions on the left hemisphere. Paired t-tests reveal improved regions with statistical significance after the FDR correction (q = 0.05). 46 and 24 out of 49 regions are improved against multi-atlas and spherical U-Net approaches, respectively. The color in the x -axis labels indicates the improved regions: multi-atlas (blue), both approaches (green), and no improvement (black). spherica… view at source ↗
read the original abstract

We present cortical surface parcellation using spherical deep convolutional neural networks. Traditional multi-atlas cortical surface parcellation requires inter-subject surface registration using geometric features with high processing time on a single subject (2-3 hours). Moreover, even optimal surface registration does not necessarily produce optimal cortical parcellation as parcel boundaries are not fully matched to the geometric features. In this context, a choice of training features is important for accurate cortical parcellation. To utilize the networks efficiently, we propose cortical parcellation-specific input data from an irregular and complicated structure of cortical surfaces. To this end, we align ground-truth cortical parcel boundaries and use their resulting deformation fields to generate new pairs of deformed geometric features and parcellation maps. To extend the capability of the networks, we then smoothly morph cortical geometric features and parcellation maps using the intermediate deformation fields. We validate our method on 427 adult brains for 49 labels. The experimental results show that our method out-performs traditional multi-atlas and naive spherical U-Net approaches, while achieving full cortical parcellation in less than a minute.

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.

Referee Report

2 major / 1 minor

Summary. The paper claims that spherical CNNs can be trained for cortical surface parcellation by generating augmented training pairs via deformation fields obtained from aligning ground-truth parcel boundaries; these fields are used to morph both geometric features and label maps. On a cohort of 427 adult brains and 49 labels, the resulting model is reported to outperform both traditional multi-atlas registration (2-3 h per subject) and a naive spherical U-Net while completing parcellation in under one minute.

Significance. If the augmentation procedure generalizes without systematic bias and the performance gains are reproducible, the work would offer a practical, high-throughput alternative to slow registration-based parcellation pipelines, which is relevant for large-scale neuroimaging studies. The empirical, held-out validation on hundreds of subjects is a positive feature of the design.

major comments (2)
  1. [Abstract] Abstract (data-augmentation paragraph): the deformation fields are derived by aligning ground-truth parcel boundaries rather than independent geometric features; this choice risks producing training distributions that do not match the geometry of unseen test subjects, so the reported superiority over multi-atlas and naive U-Net may be attributable to the augmentation rather than the network architecture itself. No description of the alignment algorithm, regularization of the fields, or quantitative checks that the morphed surfaces remain anatomically plausible is supplied.
  2. [Abstract] Validation description (Abstract): the claim of outperformance on 427 brains lacks accompanying error bars, explicit train/test split details, exclusion criteria, or statistical tests, preventing assessment of whether the gains are robust or merely reflect the particular augmentation distribution.
minor comments (1)
  1. [Abstract] The abstract states that 'a choice of training features is important' but does not enumerate which geometric features are actually supplied to the network; this omission reduces clarity of the input representation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment below and indicate planned revisions where appropriate.

read point-by-point responses

  1. Referee: [Abstract] Abstract (data-augmentation paragraph): the deformation fields are derived by aligning ground-truth parcel boundaries rather than independent geometric features; this choice risks producing training distributions that do not match the geometry of unseen test subjects, so the reported superiority over multi-atlas and naive U-Net may be attributable to the augmentation rather than the network architecture itself. No description of the alignment algorithm, regularization of the fields, or quantitative checks that the morphed surfaces remain anatomically plausible is supplied.

    Authors: The augmentation strategy is a core component of the proposed method rather than an extraneous factor; the naive spherical U-Net baseline does not employ this parcel-boundary-driven deformation augmentation, allowing the comparison to isolate its contribution alongside the spherical CNN architecture. The held-out test performance on 427 subjects demonstrates generalization to unseen geometries. We agree that the abstract lacks methodological specifics and will expand the methods section in revision to describe the alignment algorithm, any regularization applied to the deformation fields, and quantitative checks confirming that morphed surfaces remain anatomically plausible. revision: yes

  2. Referee: [Abstract] Validation description (Abstract): the claim of outperformance on 427 brains lacks accompanying error bars, explicit train/test split details, exclusion criteria, or statistical tests, preventing assessment of whether the gains are robust or merely reflect the particular augmentation distribution.

    Authors: The full manuscript contains the train/test split, exclusion criteria, and dataset description. We acknowledge that the abstract is too concise on these points and omits error bars or statistical tests. In revision we will update the abstract to reference the cross-validation procedure and report mean performance with standard deviations; we will also ensure the results section includes appropriate statistical comparisons. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical method with held-out validation

full rationale

The paper describes a data-driven spherical CNN pipeline for cortical parcellation that augments training data via deformation fields obtained from ground-truth boundary alignments and reports performance on a held-out set of 427 brains. No equations, predictions, or uniqueness claims reduce the reported outperformance to fitted inputs or self-citation chains by construction. The central result is an empirical comparison against multi-atlas and naive U-Net baselines on independent test subjects, making the derivation self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Review based on abstract only; no explicit free parameters, axioms, or invented entities are stated. Implicit assumptions include standard CNN training convergence and that spherical representation preserves relevant cortical geometry.

pith-pipeline@v0.9.0 · 5775 in / 1023 out tokens · 27788 ms · 2026-05-24T22:37:30.087348+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

16 extracted references · 16 canonical work pages · 3 internal anchors

  1. [1]

    Journal of the Royal statistical society: series B (Methodological) 57(1), 289–300 (1995)

    Benjamini, Y., Hochberg, Y.: Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal statistical society: series B (Methodological) 57(1), 289–300 (1995)

    work page 1995

  2. [2]

    IEEE Transactions on Pattern Analysis & Machine Intelligence 26(9), 1124–1137 (2004)

    Boykov, Y., Kolmogorov, V.: An experimental comparison of min-cut/max-flow al- gorithms for energy minimization in vision. IEEE Transactions on Pattern Analysis & Machine Intelligence 26(9), 1124–1137 (2004)

    work page 2004

  3. [3]

    Spherical CNNs

    Cohen, T.S., Geiger, M., K¨ ohler, J., Welling, M.: Spherical cnns. arXiv preprint arXiv:1801.10130 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  4. [4]

    In: International Conference on Medical Imaging with Deep Learning (2018)

    Cucurull, G., Wagstyl, K., Casanova, A., Veliˇ ckovi´ c, P., Jakobsen, E., Drozdzal, M., Romero, A., Evans, A., Bengio, Y.: Convolutional neural networks for mesh- based parcellation of the cerebral cortex. In: International Conference on Medical Imaging with Deep Learning (2018)

    work page 2018

  5. [5]

    Neuroimage 31(3), 968–980 (2006)

    Desikan, R.S., S´ egonne, F., Fischl, B., Quinn, B.T., Dickerson, B.C., Blacker, D., Buckner, R.L., Dale, A.M., Maguire, R.P., Hyman, B.T., et al.: An automated labeling system for subdividing the human cerebral cortex on mri scans into gyral based regions of interest. Neuroimage 31(3), 968–980 (2006)

    work page 2006

  6. [6]

    In: European Conference on Computer Vision (2018)

    Esteves, C., Allen-Blanchette, C., Makadia, A., Daniilidis, K.: Learning so(3) equiv- ariant representations with spherical cnns. In: European Conference on Computer Vision (2018)

    work page 2018

  7. [7]

    Cerebral cortex 14(1), 11–22 (2004)

    Fischl, B., Van Der Kouwe, A., Destrieux, C., Halgren, E., S´ egonne, F., Salat, D.H., Busa, E., Seidman, L.J., Goldstein, J., Kennedy, D., et al.: Automatically parcellating the human cerebral cortex. Cerebral cortex 14(1), 11–22 (2004)

    work page 2004

  8. [8]

    Graph Convolutions on Spectral Embeddings: Learning of Cortical Surface Data

    Gopinath, K., Desrosiers, C., Lombaert, H.: Graph convolutions on spectral em- beddings: Learning of cortical surface data. arXiv preprint arXiv:1803.10336 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  9. [9]

    Spherical CNNs on Unstructured Grids

    Jiang, C., Huang, J., Kashinath, K., Marcus, P., Niessner, M., et al.: Spherical cnns on unstructured grids. arXiv preprint arXiv:1901.02039 (2019)

    work page internal anchor Pith review Pith/arXiv arXiv 1901

  10. [10]

    In: 16th Annual Meeting for the Organization of Human Brain Mapping (2010) Cortical Surface Parcellation using Spherical Convolutional Neural Networks 9

    Klein, A., Dal Canton, T., Ghosh, S.S., Landman, B., Lee, J., Worth, A.: Open labels: online feedback for a public resource of manually labeled brain images. In: 16th Annual Meeting for the Organization of Human Brain Mapping (2010) Cortical Surface Parcellation using Spherical Convolutional Neural Networks 9

    work page 2010

  11. [11]

    Neuroimage 34(4), 1535–1544 (2007)

    Lyttelton, O., Boucher, M., Robbins, S., Evans, A.: An unbiased iterative group registration template for cortical surface analysis. Neuroimage 34(4), 1535–1544 (2007)

    work page 2007

  12. [12]

    Medical image analysis 57, 72–88 (2019)

    Lyu, I., Kang, H., Woodward, N.D., Styner, M.A., Landman, B.A.: Hierarchical spherical deformation for cortical surface registration. Medical image analysis 57, 72–88 (2019)

    work page 2019

  13. [13]

    Medical image analysis 48, 244–258 (2018)

    Lyu, I., Kim, S.H., Girault, J.B., Gilmore, J.H., Styner, M.A.: A cortical shape- adaptive approach to local gyrification index. Medical image analysis 48, 244–258 (2018)

    work page 2018

  14. [14]

    SIAM Journal on Numerical Analysis 41(1), 325–363 (2003)

    Sethian, J.A., Vladimirsky, A.: Ordered upwind methods for static hamilton–jacobi equations: Theory and algorithms. SIAM Journal on Numerical Analysis 41(1), 325–363 (2003)

    work page 2003

  15. [15]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Wu, Z., Li, G., Wang, L., Shi, F., Lin, W., Gilmore, J.H., Shen, D.: Registration- free infant cortical surface parcellation using deep convolutional neural networks. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 672–680. Springer (2018)

    work page 2018

  16. [16]

    Journal of neurophysiology 106(3), 1125 (2011)

    Yeo, B.T., Krienen, F.M., Sepulcre, J., Sabuncu, M.R., Lashkari, D., Hollinshead, M., Roffman, J.L., Smoller, J.W., Z¨ ollei, L., Polimeni, J.R., et al.: The organization of the human cerebral cortex estimated by intrinsic functional connectivity. Journal of neurophysiology 106(3), 1125 (2011)

    work page 2011