REVIEW 3 major objections 5 minor 23 references
Fully Automated Segmentation of Fiber Bundles in Anatomic Tracing Data
T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash
Pith's one-line read This paper claims that a U-Net framework can automatically segment fiber bundles in macaque tracer data, beating prior methods on sparse bundle detection and false discovery rate while working on single slides.
desk verdict Promising and honest methods preprint for automated macaque tracer segmentation, but the version we have omits the entire evaluation, so the headline gains are unverifiable and should not be cited yet. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central machinery is a U-Net, a convolutional encoder–decoder that maps an image patch to a pixel-wise segmentation map, configured with large patch sizes so that each prediction has surrounding anatomical context. Foreground-aware sampling preferentially selects patches containing tracer signal, countering the class imbalance in which sparse fiber bundles occupy few pixels. Semi-supervised pre-training on unlabeled sections stretches the limited manual annotations before the final supervised fine-tuning. Together these components do the work of separating true fiber bundles from terminal boutons and background noise.
What would settle it
On a held-out set of sections with separate expert labels for terminals and bundles, count how many terminal voxels the model predicts as bundles; the claim that terminal mislabeling is eliminated falls if that fraction is not near zero, and the 40% false-discovery-rate reduction falls if the comparison against the prior method on the same single slices does not reproduce.
Extended reading notes
Core claim
On the paper's own terms, the central discovery is that a U-Net with large input patches, foreground-aware sampling, and semi-supervised pre-training is enough to segment fiber bundles in macaque anatomic tracer histology accurately enough to outperform the prior deep-learning method. The authors state that their model avoids the common error of labeling terminal boutons as fiber bundles, detects sparse bundles over 20% better, and lowers the false discovery rate by 40%, all without the neighbor-aware post-processing that limited earlier work to full series of consecutive sections. The method is evaluated across multiple macaque brains with tracer injections at different sites, supporting the claim of generalization across brains and fiber configurations. The paper's limitations section notes that additional object- and pixel-level metrics and variance across runs are planned future work, so the reported gains are tied to the evaluation metrics used.
Load-bearing premise
The load-bearing premise is that fiber bundles and terminal boutons look different enough in tracer-stained sections for a model trained on expert labels to learn the distinction; if they look alike under the microscope, the reported elimination of terminal mislabeling cannot hold.
Editorial extensions
If this is right
- Automated segmentation can produce ground-truth bundle annotations across many macaque brains without manual tracing, removing the annotation bottleneck that limits validation studies of diffusion MRI tractography.
- Because the method works on single sections, it can be applied to archival or partial slide collections where consecutive-section context is unavailable.
- Lower false discovery rates mean tractography comparisons will include fewer terminal boutons mislabeled as long-range bundles, making downstream validation less noisy.
- The framework provides a fully automated baseline, so future attempts to segment tracer data no longer need complex cross-section post-processing.
Reading between the lines
- If the same recipe transfers to other primates or to human post-mortem tissue, the method could become a general digitized-anatomy tool rather than a macaque-specific one.
- The paper leaves implicit which component contributes most of the sparse-bundle gain; an ablation of large patches versus foreground-aware sampling versus pre-training would isolate the mechanism.
- The authors' planned 3D extension suggests the 2D gains could carry over to volumetric light-sheet fluorescence microscopy, where adjacent-section context is intrinsically available.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript proposes a semi-supervised U-Net framework for fully automated segmentation of fiber bundles in macaque anatomic tracer data, using large patch sizes, foreground-aware sampling, and semi-supervised pre-training. The abstract and introduction make strong quantitative claims: a 22% improvement in detecting sparse fiber bundles, a 40% reduction in false discovery rate (FDR), and the elimination of errors that mislabel terminals as bundles, all while enabling analysis of standalone slices. A public GitHub repository is provided. The submitted text, however, contains only the abstract, introduction, a short future-work paragraph, and conclusions; the sections describing the method and the experimental evaluation (Sections 2–5) are missing, and the future-work paragraph explicitly states that pixel-level metrics (Dice, IoU, detection rate) and variance across runs are not yet reported.
Significance. If the claimed results hold, the framework would be a practical step toward automating the analysis of anatomic tracer studies, generating much-needed ground-truth data for validating dMRI tractography. The public code release, the use of semi-supervised learning to address limited manual annotations, and the goal of single-section analysis are all strengths. However, the significance of the paper as currently submitted cannot be assessed because the central quantitative claims rest entirely on an evaluation that is not present in the manuscript. The claims are falsifiable and the code is available, so the work is testable, but the evidence as presented is incomplete.
major comments (3)
- [Abstract and future-work paragraph] The manuscript omits the entire methods and evaluation sections (Sections 2–5), yet the abstract and introduction assert a 22% improvement in sparse-bundle detection and a 40% reduction in FDR. The future-work paragraph explicitly acknowledges that object- and pixel-level metrics (Dice, IoU, detection rate) and variance across runs are not yet reported. Without a description of the dataset, training protocol, baseline implementation, and the evaluation procedure, these headline quantitative claims are unverifiable in the submitted text. This is load-bearing because the paper's central contribution is empirical.
- [Abstract, Section 1, future-work paragraph] The 40% FDR reduction is not a controlled comparison as described. FDR is an object-level precision-like metric, and the manuscript gives no information about how the baseline of [16] was thresholded, whether the baseline's multi-section post-processing was applied, or whether the new single-section setting changes the operating point. The future-work admission that detection rate is missing leaves open the possibility that the FDR improvement is a sensitivity trade-off rather than a genuine error reduction. The authors should report precision-recall curves or detection-rate results at matched operating points to support the claim.
- [Section 1 and future-work paragraph] The claim that the method 'eliminates' mislabeling of terminals as bundles is categorical and not supported by any error analysis in the provided text. The implicit premise that terminal boutons and fiber bundles are visually distinguishable in the tracer-stained images is also not demonstrated. Since the future-work paragraph indicates that detection rate and pixel-level metrics are still missing, the current manuscript does not establish that the model detects all relevant bundles, nor that the terminal/bundle separation is achieved. An error analysis breaking down false positives by type (terminals vs. other structures) is needed.
minor comments (5)
- [Abstract vs. Introduction] The abstract reports an improvement of 'over 20%' for sparse bundles, while the Introduction states a more specific '22% improvement'; these numbers should be reconciled or the source clarified.
- [Abstract] The abstract introduces FDR without defining it or specifying how it is computed at the object level; please define the metric and describe the matching criterion for detections.
- [Page 10 (future-work paragraph)] The future-work paragraph appears to be part of a section whose heading is missing from the provided text; ensure all section headings are present in the final submission.
- [References] Reference [16] is cited as 'Imaging Neuroscience 3, imag_a_00514 (2025)'; please verify the formatting, volume/page details, and ensure all references are complete.
- [GitHub link] The GitHub link is a welcome resource, but it does not substitute for a methods section; the manuscript should describe the architecture, sampling scheme, and training procedure in sufficient detail for reproducibility.
Circularity Check
No significant circularity: the claims are empirical comparisons against a published baseline; the missing methods and experiments sections are a completeness gap, not a circular derivation.
full rationale
The visible text contains no derivation chain in which a predicted quantity is equivalent to an input by construction. The central quantitative claims, sparse-bundle improvement and FDR reduction, are presented as measured outcomes against the state-of-the-art method [16], which is a separately published baseline even though it shares research-group members. That shared provenance is a reason to inspect the evaluation protocol, but it does not make the comparison circular: the prior method was externally published and the current code is public, so the comparison is in principle falsifiable. The paper's own future-work note, 'we aim to include additional object- and pixel-level metrics (e.g., Dice coefficient, IoU, detection rate) in future work,' admits that pixel-level metrics and variance are missing; this weakens the evidential strength of the headline numbers but is a reporting limitation, not a reduction of the result to the input. No fitted parameter is relabeled as a prediction, no uniqueness theorem is imported from the authors' earlier work, and no ansatz is smuggled in via self-citation. The omission of Sections 2-5 from the provided text prevents independent recomputation of the metrics, but auditability gaps are correctness risks, not circularity.
Assumptions & free parameters
assumptions (3)
- domain assumption Manual expert annotations of fiber bundles on histological slides are accurate ground truth.
- domain assumption Tracer-stained fiber bundles and terminal boutons have distinguishable image appearance.
- domain assumption Semisupervised pretraining on unlabeled data improves segmentation performance without introducing systematic bias.
Cite this review
Pith. "Pith review of Fully Automated Segmentation of Fiber Bundles in Anatomic Tracing Data." pith.science (2026). https://pith.science/paper/IN7L5EAF
@misc{pith2026250812942,
author = {Pith},
title = {Pith review of: Fully Automated Segmentation of Fiber Bundles in Anatomic Tracing Data},
year = {2026},
howpublished = {\url{https://pith.science/paper/IN7L5EAF}},
note = {Machine review of arXiv:2508.12942}
}
read the original abstract
Anatomic tracer studies are critical for validating and improving diffusion MRI (dMRI) tractography. However, large-scale analysis of data from such studies is hampered by the labor-intensive process of annotating fiber bundles manually on histological slides. Existing automated methods often miss sparse bundles or require complex post-processing across consecutive sections, limiting their flexibility and generalizability. We present a streamlined, fully automated framework for fiber bundle segmentation in macaque tracer data, based on a U-Net architecture with large patch sizes, foreground aware sampling, and semisupervised pre-training. Our approach eliminates common errors such as mislabeling terminals as bundles, improves detection of sparse bundles by over 20% and reduces the False Discovery Rate (FDR) by 40% compared to the state-of-the-art, all while enabling analysis of standalone slices. This new framework will facilitate the automated analysis of anatomic tracing data at a large scale, generating more ground-truth data that can be used to validate and optimize dMRI tractography methods.
Reference graph
Works this paper leans on
-
[16]
Imag- ing Neuroscience 3, imag_a_00514 (2025)
Sundaresan, V., Lehman, J.F., Maffei, C., Haber, S.N., Yendiki, A.: Self-supervised segmentation and characterization of fiber bundles in anatomic tracing data. Imag- ing Neuroscience 3, imag_a_00514 (2025)
work page 2025
-
[1]
Current opinion in neurology19(6), 599–606 (2006)
Catani, M.: Diffusion tensor magnetic resonance imaging tractography in cognitive disorders. Current opinion in neurology19(6), 599–606 (2006)
work page 2006
-
[2]
BMC evolutionary biology 9, 1–19 (2009)
Chatterjee, H.J., Ho, S.Y., Barnes, I., Groves, C.: Estimating the phylogeny and divergence times of primates using a supermatrix approach. BMC evolutionary biology 9, 1–19 (2009)
work page 2009
-
[3]
Proceedings of the National Academy of Sciences117(20), 11068–11075 (2020)
Friedmann, D., Pun, A., Adams, E.L., Lui, J.H., Kebschull, J.M., Grutzner, S.M., Castagnola, C., Tessier-Lavigne, M., Luo, L.: Mapping mesoscale axonal projec- tions in the mouse brain using a 3d convolutional network. Proceedings of the National Academy of Sciences117(20), 11068–11075 (2020)
work page 2020
-
[4]
Grisot, G., Haber, S.N., Yendiki, A.: Diffusion mri and anatomic tracing in the same brain reveal common failure modes of tractography. Neuroimage239, 118300 (2021) Fully Automated Segmentation of Fiber Bundles in Anatomic Tracing Data 11
work page 2021
-
[5]
Journal of Neuroscience 33(11), 4804–4814 (2013)
Haynes, W.I., Haber, S.N.: The organization of prefrontal-subthalamic inputs in primates provides an anatomical substrate for both functional specificity and inte- gration: implications for basal ganglia models and deep brain stimulation. Journal of Neuroscience 33(11), 4804–4814 (2013)
work page 2013
-
[6]
Journal of Neuroscience33(7), 3190–3201 (2013)
Jbabdi, S., Lehman, J.F., Haber, S.N., Behrens, T.E.: Human and monkey ven- tral prefrontal fibers use the same organizational principles to reach their targets: tracing versus tractography. Journal of Neuroscience33(7), 3190–3201 (2013)
work page 2013
-
[7]
arXiv preprint arXiv:1412.6980 (2014)
Kingma, D.P., Ba, J.: Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 (2014)
arXiv 2014
Show all 23 references
-
[8]
Journal of Neuroscience31(28), 10392–10402 (2011)
Lehman, J.F., Greenberg, B.D., McIntyre, C.C., Rasmussen, S.A., Haber, S.N.: Rules ventral prefrontal cortical axons use to reach their targets: implications for diffusion tensor imaging tractography and deep brain stimulation for psychiatric illness. Journal of Neuroscience31...
2011
-
[9]
Lin, T.Y., Goyal, P., Girshick, R., He, K., Dollár, P.: Focal loss for dense object detection.In:ProceedingsoftheIEEEinternationalconferenceoncomputervision. pp. 2980–2988 (2017)
2017
-
[10]
Journal of neuroscience methods291, 141–149 (2017)
Naito, T., Nagashima, Y., Taira, K., Uchio, N., Tsuji, S., Shimizu, J.: Identification and segmentation of myelinated nerve fibers in a cross-sectional optical microscopic image using a deep learning model. Journal of neuroscience methods291, 141–149 (2017)
2017
-
[11]
arXiv preprint arXiv:1912.01703 (2019)
Paszke, A.: Pytorch: An imperative style, high-performance deep learning library. arXiv preprint arXiv:1912.01703 (2019)
2019 arXiv
-
[12]
In: Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, Oc- tober 5-9, 2015, proceedings, part III 18
Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomed- ical image segmentation. In: Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, Oc- tober 5-9, 2015, proceedings, part III 18....
2015
-
[13]
Journal of Neuroscience 38(8), 2106–2117 (2018)
Safadi, Z., Grisot, G., Jbabdi, S., Behrens, T.E., Heilbronner, S.R., McLaughlin, N.C., Mandeville, J., Versace, A., Phillips, M.L., Lehman, J.F., et al.: Functional segmentation of the anterior limb of the internal capsule: linking white matter abnormalities to specific conne...
2018
-
[14]
Neuroimage185, 1–11 (2019)
Schilling, K.G., Nath, V., Hansen, C., Parvathaneni, P., Blaber, J., Gao, Y., Neher, P., Aydogan, D.B., Shi, Y., Ocampo-Pineda, M., et al.: Limits to anatomical ac- curacy of diffusion tractography using modern approaches. Neuroimage185, 1–11 (2019)
2019
-
[15]
In: International Workshop on Medical Optical Imaging and Virtual Microscopy Image Analysis
Sundaresan, V., Lehman, J.F., Fitzgibbon, S., Jbabdi, S., Haber, S.N., Yendiki, A.: Constrained self-supervised method with temporal ensembling for fiber bundle detection on anatomic tracing data. In: International Workshop on Medical Optical Imaging and Virtual Microscopy Ima...
2022
-
[17]
Proceedings of the National Academy of Sciences 111(46), 16574–16579 (2014)
Thomas, C., Ye, F.Q., Irfanoglu, M.O., Modi, P., Saleem, K.S., Leopold, D.A., Pierpaoli, C.: Anatomical accuracy of brain connections derived from diffusion mri tractography is inherently limited. Proceedings of the National Academy of Sciences 111(46), 16574–16579 (2014)
2014
-
[18]
Wei, D., Lee, K., Li, H., Lu, R., Bae, J.A., Liu, Z., Zhang, L., dos Santos, M., Lin, Z., Uram, T., et al.: Axonem dataset: 3d axon instance segmentation of brain cor- tical regions. In: Medical Image Computing and Computer Assisted Intervention– MICCAI 2021: 24th Internationa...
2021
-
[19]
Cell179(1), 268–281 (2019)
Winnubst, J., Bas, E., Ferreira, T.A., Wu, Z., Economo, M.N., Edson, P., Arthur, B.J.,Bruns,C.,Rokicki,K.,Schauder,D.,etal.:Reconstructionof1,000projection neurons reveals new cell types and organization of long-range connectivity in the mouse brain. Cell179(1), 268–281 (2019)
2019
-
[20]
Brain Structure and Function225, 1225–1243 (2020)
Woodward, A., Gong, R., Abe, H., Nakae, K., Hata, J., Skibbe, H., Yamaguchi, Y., Ishii, S., Okano, H., Yamamori, T., et al.: The nanozoomer artificial intelligence connectomics pipeline for tracer injection studies of the marmoset brain. Brain Structure and Function225, 1225–1...
2020
-
[21]
Elife 11, e72534 (2022)
Yan, M., Yu, W., Lv, Q., Lv, Q., Bo, T., Chen, X., Liu, Y., Zhan, Y., Yan, S., Shen, X., et al.: Mapping brain-wide excitatory projectome of primate prefrontal cortex at submicron resolution and comparison with diffusion tractography. Elife 11, e72534 (2022)
2022
-
[22]
Physics in Medicine & Biology66(15), 15TR01 (2021)
Yang, J.Y.M., Yeh, C.H., Poupon, C., Calamante, F.: Diffusion mri tractogra- phy for neurosurgery: the basics, current state, technical reliability and challenges. Physics in Medicine & Biology66(15), 15TR01 (2021)
2021
-
[23]
Neuroimage 256, 119146 (2022)
Yendiki, A., Aggarwal, M., Axer, M., Howard, A.F., van Walsum, A.M.v.C., Haber, S.N.: Post mortem mapping of connectional anatomy for the validation of diffusion mri. Neuroimage 256, 119146 (2022)
2022
Reviewed August 15, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.