{"id":"9aafa0a2-380e-47db-a113-4384729af97b","arxiv_id":"2508.12942","paper_version":2,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A U-Net pipeline with large patches, foreground aware sampling, and semisupervised pretraining automatically segments fiber bundles in macaque tracer data, improving sparse bundle detection and reducing false positives.","lead":"This paper describes an automated deep-learning method for segmenting fiber bundles in microscope images of macaque brains with tracer injections. It claims a 20% gain in sparse bundle detection and a 40% cut in false positives over prior work, while working on individual tissue sections.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 40% FDR and 20% sparse-bundle gains are unverifiable because the methods/evaluation sections are missing and the paper's own future-work note admits no pixel-level metrics or variance; the key unaddressed risk is that the FDR gain reflects an uncontrolled operating-point or sensitivity…","rationale":"The task is to stress-test the central claim: a U-Net-based framework that improves sparse-bundle detection by more than 20% and reduces FDR by 40% versus the state of the art, while working on single slices. I read the available text in good faith: the introduction frames the problem well, the comparison to prior work by Sundaresan et al. is plausible, and a public GitHub link is a genuine asset. The central claim, however, rests entirely on Sections 2–5, which are absent from the provided text. We have only the abstract, introduction, a short future-work paragraph, and conclusions. That absence alone makes the results unverifiable; the future-work paragraph then tells us what the evaluation does not contain: Dice, IoU, detection rate, and variance across runs. These are not cosmetic additions. Without a detection-rate or recall figure, an FDR reduction is compatible with the model simply predicting fewer bundles overall. Without variance or a matched operating point, the sparse-bundle improvement cannot be separated from threshold effects. The reader's weakest assumption about visual distinguishability of terminals and bundles is a real precondition, but the more directly load-bearing gap is the uncontrolled object-level metric comparison. I would not call the paper wrong; the available evidence is simply insufficient. Therefore the verdict remains UNVERDICTED. If the full paper or the public code reproduces the numbers at matched recall with confidence intervals, the central claim would be substantially supported.","tokens_in":4088,"tokens_out":3633,"duration_ms":38586,"concrete_test":"Retrieve the missing Sections 2–5 or run the public GitHub code on the same dataset, then reproduce the FDR and sparse-bundle detection numbers under three conditions: (1) with a shared detection threshold or matched operating point across the new method and the Sundaresan et al. baseline; (2) with recall or detection rate reported alongside FDR so a precision-recall trade-off is visible; and (3) with multiple training runs or held-out brains to obtain variance. If the 40% FDR reduction persists at matched recall with overlapping confidence intervals, the claim holds; if it shrinks or disappears, the abstract overstates the result.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The provided manuscript contains the abstract, introduction, a future-work paragraph, and conclusions, but omits the methods and experimental sections (Sections 2–5) that would define the evaluation. The strongest quantitative claims—>20% improvement on sparse bundles, 40% FDR reduction, and 'eliminating' terminal mislabeling—therefore rest on an unseen comparison protocol. The most load-bearing unverified point is that the reported FDR reduction reflects a genuine improvement rather than a shift in operating point or a sensitivity trade-off: FDR is a precision-like object-level metric, and if the baseline is evaluated with its own threshold while the new method uses a different one, or if the baseline's multi-section post-processing is not matched to the new single-section setting, the comparison is not controlled. The paper itself notes in future work that pixel-level metrics (Dice, IoU, detection rate) and variance across runs are still missing, which means the current report does not establish that the model detects all relevant bundles, only that its positive predictions are more often correct under an unspecified criterion. Since the GitHub code is public, this gap is testable rather than fatal, but until the evaluation protocol is available the headline numbers cannot be independently interpreted.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","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.","tokens_in":4321,"tokens_out":2582,"duration_ms":26234,"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":[{"comment":"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.","section":"Abstract and future-work paragraph"},{"comment":"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":"Abstract, Section 1, future-work paragraph"},{"comment":"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.","section":"Section 1 and future-work paragraph"}],"minor_comments":[{"comment":"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.","section":"Abstract vs. Introduction"},{"comment":"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.","section":"Abstract"},{"comment":"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.","section":"Page 10 (future-work paragraph)"},{"comment":"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.","section":"References"},{"comment":"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.","section":"GitHub link"}],"recommendation":"major_revision","confidential_remarks":"The submitted text is materially incomplete: Sections 2–5, which would contain the method and experimental results, are absent. The strong quantitative claims in the abstract and introduction cannot be evaluated without them. If the authors have a complete version under review elsewhere, they should provide it; otherwise, the manuscript needs substantial additions. I recommend major revision rather than rejection because the core idea is plausible and testable, and the missing evaluation is a fixable gap within the paper's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the idea is solid and the code is public, but the draft we have is missing the methods and experimental sections, so the 20% and 40% numbers are promises, not results. The paper's own future-work paragraph admits that pixel-level metrics (Dice, IoU, detection rate) and variance across runs are not yet reported. If you rely on ground truth for tractography validation, watch the repo, but don't cite the numbers yet.\n\nWhat's actually new: applying a U-Net with large patches, foreground-aware sampling, and semi-supervised pretraining to macaque anterograde tracer sections, with the explicit goal of detecting sparse bundles and suppressing terminal boutons. That is a real gap: the prior state of the art [16] needed multi-section post-processing and missed sparse bundles. Automating this annotation step matters for scaling ground-truth generation. The code is on GitHub, which makes the claims testable.\n\nWhat's good: the problem is well motivated, the design choices are reasonable, and the comparison is against the same group's earlier method, which is legitimate, not circular. The writing is honest about limitations—the future-work passage openly says Dice/IoU/detection rate and variance are still missing. That honesty works in its favor.\n\nWhere it's soft: the abstract's \"over 20%\" and \"40% FDR reduction\" cannot be checked because Sections 2–5 are absent. The one number we do see in the intro, 22%, does not match the abstract's wording. More importantly, FDR is an operating-point-dependent object-level metric. Without the baseline threshold, matching criterion, and whether the comparison is single- or multi-section, the reported improvement could reflect a sensitivity trade-off rather than a genuine gain. The paper's own admission that detection rate and variance are missing means we don't know whether the model finds all relevant bundles or just behaves well on a subset. This is a load-bearing gap, not a cosmetic one.\n\nThis paper is for neuroimagers working on tractography validation and connectomics scale-up. It would be a useful contribution once the full evaluation is released. I would send it to peer review with a clear request for major revision: restore the methods/experiments, add pixel-level metrics and variance, and show operating-point comparisons. As it stands, bring it to reading group for discussion of incomplete preprints, but don't cite it and don't treat the headline numbers as established.","headline":"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.","tokens_in":4858,"tokens_out":2543,"would_cite":false,"duration_ms":27745,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"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.","keywords":["fiber bundle segmentation","anatomic tracing","U-Net","semi-supervised learning","false discovery rate","sparse bundle detection","histological image segmentation","macaque brain"],"falsifier":"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.","tokens_in":3911,"feed_emoji":"🧠","tokens_out":11103,"duration_ms":100610,"temperature":0.7,"pith_summary":"Anatomic tracer studies make the axonal projections of injected brain regions visible in microscope slides, giving diffusion MRI tractography a direct ground truth to check against. The bottleneck is that labeling the fiber bundles on each slide by hand is slow and labor-intensive. This paper claims that a fully automated U-Net, trained with foreground-aware sampling and semi-supervised pre-training, can segment fiber bundles in macaque tracer data without needing post-processing across consecutive sections. The reported gains over the previous state of the art are over 20% for sparse bundle detection and a 40% reduction in false discovery rate, with the method working on single slides. If the claim holds, large-scale extraction of ground-truth annotations for tractography validation becomes practical.","feed_headline":"Automated fiber-bundle segmentation cuts false positives by 40%","feed_subtitle":"U-Net with foreground-aware sampling beats prior methods on sparse bundles and works on single slides.","key_machinery":"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.","core_discovery":"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.","pith_inferences":["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."],"forward_implications":["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."],"supporting_citations":[{"why":"The prior deep-learning method for macaque tracer segmentation that this paper's results are compared against.","marker":"[16]"},{"why":"Earlier self-supervised fiber-bundle detection work that motivates the semi-supervised pre-training strategy.","marker":"[15]"},{"why":"Introduces the U-Net architecture that the proposed segmentation framework is built on.","marker":"[12]"},{"why":"Focal loss, used as the training loss to handle the class imbalance created by sparse fiber bundles.","marker":"[9]"}],"fun_headline_variants":["Sparse fiber bundles: +20% recall, 40% fewer false hits","U-Net automates fiber tracing, sees sparse bundles 20% better","No more slide-by-slide: one U-Net slices fiber bundles","Fiber bundle mapping goes fully auto, FDR down 40%","Tracer data segmentation: better sparse detection, fewer false positives"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"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.","fun_headline_variants_meta":{"raw":{"variants":["Sparse fiber bundles: +20% recall, 40% fewer false hits","U-Net automates fiber tracing, sees sparse bundles 20% better","No more slide-by-slide: one U-Net slices fiber bundles","Fiber bundle mapping goes fully auto, FDR down 40%","Tracer data segmentation: better sparse detection, fewer false positives"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000946,"raw_usage":{"total_tokens":4009,"prompt_tokens":881,"completion_tokens":3128,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":497,"completion_tokens_details":{"reasoning_tokens":3031}},"tokens_in":497,"tokens_out":3128,"duration_ms":19137,"temperature":1.0,"reasoning_tokens":3031,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T17:15:52.536327+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"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.","supporting_citations":[{"cited_title":"Imag- ing Neuroscience 3, imag_a_00514 (2025)","cited_arxiv_id":null,"evidence_quote":"The prior deep-learning method for macaque tracer segmentation that this paper's results are compared against."},{"cited_title":"In: International Workshop on Medical Optical Imaging and Virtual Microscopy Image Analysis","cited_arxiv_id":null,"evidence_quote":"Earlier self-supervised fiber-bundle detection work that motivates the semi-supervised pre-training strategy."}],"review_version":1}