Pith. sign in

REVIEW 1 major objections 7 minor 25 references

Centre maps and synthetic mimics cut microbleed detection errors

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · glm-5.2

2026-07-07 17:05 UTC pith:XAFN265M

load-bearing objection Solid engineering for CMB detection; ablations are underpowered but the external AIBL result carries real weight. the 1 major comments →

arxiv 2607.05325 v1 pith:XAFN265M submitted 2026-07-06 cs.CV

CenSynCMB: Centre Maps and Physics-Guided Synthesis for Microbleed Detection

Lucas He , Hanyuan Zhang , Krinos Li , Adama Fatima Saccoh , Silvia Ingala , Rafael Rehwald , Marleen de Bruijne , Frederik Barkhof
show 2 more authors
Rhodri Davies Carole H. Sudre
This is my paper
classification cs.CV
keywords censyncmbachievedcmbsdetectionlocal-comparisonmicrobleedphysics-guidedsmall
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

CenSynCMB argues that two design choices fix the main failure modes of automated cerebral microbleed (CMB) detection: (1) an auxiliary centre-map head that trains the network to localise lesion centroids rather than voxel-level segmentation overlap, and (2) fold-wise physics-guided synthesis that inserts synthetic positive CMBs and labelled hard-negative mimics (vessel-like and calcification-like structures) into real MRI backgrounds. The centre-map head aligns the training signal with the centroid-matching criterion used in clinical evaluation, while the synthesis module supplies controlled negative evidence for the structures that most often generate false positives. Together, these mechanisms are shown to improve lesion-level F1 on the VALDO benchmark (74.3%, p=0.020) and on the external AIBL SWI cohort (65.0%, p=0.0016), where the model must transfer from T2*-GRE to susceptibility-weighted imaging. The external gains come with higher false-positive burden, and patient-level count calibration remains unreliable, so the authors frame the system as a high-throughput candidate-extraction layer rather than a finished burden-estimation tool.

Core claim

The paper's central claim is that coupling explicit centroid supervision with physics-guided synthesis of both positive lesions and labelled hard-negative mimics yields a better balance of recall and precision for CMB detection than segmentation-backbone or cascaded-detector approaches, across both in-domain and cross-sequence MRI settings. The centre-map head shifts the training objective toward the detection target (lesion centroids), and the synthesis module provides the scarce negative examples (vessel cross-sections, calcification-like foci) that real datasets lack. The combination improves lesion-level F1 on both VALDO and external AIBL, though the external recall gains are partially抵消

What carries the argument

Centre-map supervision (Gaussian target around each lesion centroid), false-negative-driven crop reweighting (Tversky loss with α=0.1, β=0.9, plus per-lesion crop weights), and physics-guided synthesis (static-dephasing susceptibility model with dipole kernel convolution for positive CMBs; elongated anisotropic masks for vessel-like mimics; compact low-signal sources for calcification-like mimics).

Load-bearing premise

The physics-guided synthesis module uses a magnitude-only static-dephasing susceptibility approximation that cannot distinguish calcification from blood products by susceptibility sign, and the synthesis ablation did not reach statistical significance (p>0.05 for all comparisons), so the claim that synthetic mimics meaningfully reduce false positives rests on trend-level directional evidence only.

What would settle it

If the synthetic mimics do not adequately represent the real false-positive distribution in external cohorts, the external recall gains on AIBL could be partially offset by the higher false-positive burden, undermining the practical utility of the detection improvement for burden estimation.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • CMB candidate extraction at scale becomes feasible for large unlabelled MRI cohorts, enabling semi-automated triage where a sensitive detector flags candidates and a human reader confirms or rejects them.
  • The centre-map supervision principle could transfer to other compact, sparse lesion types in neuroimaging (e.g., lacunes, enlarged perivascular spaces) where the evaluation target is a centroid rather than a segmentation mask.
  • The gap between lesion-level detection gains and patient-level burden calibration suggests that downstream cohort studies will need cohort-specific threshold tuning or uncertainty-aware review layers before using detector outputs as clinical counts.
  • The physics-guided synthesis approach could be extended to other susceptibility-sensitive markers or to phase/QSM-based rendering if the magnitude-only limitation is addressed.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

1 major / 7 minor

Summary. This paper proposes CenSynCMB, a centre-guided CMB detection framework combining an Attention U-Net backbone with auxiliary centre-map supervision, false-negative-driven reweighting, and fold-wise physics-guided synthesis of positive CMBs and labelled hard negatives (vessel-like and calcification-like mimics). The system is evaluated on VALDO Task 2 (5-fold CV, n=72) and externally on AIBL SWI (n=370) under a shared centroid-matching protocol with bootstrap CIs and paired Wilcoxon tests. The central claims are that CenSynCMB achieves the best local-comparison lesion-level F1 on VALDO (74.3%, p=0.020) and the highest local recall and F1 on AIBL (88.5%, p=0.0058; 65.0%, p=0.0016), supporting scalable CMB candidate extraction across GRE-to-SWI shifts. The evaluation protocol is clearly defined, comparison methods are appropriately separated into leakage-free reimplemented models and pretrained references, and synthesis priors are estimated only from training-fold subjects.

Significance. The paper addresses a clinically relevant problem (automated CMB detection for CSVD and ARIA-H monitoring) with a well-motivated design. Strengths include: (1) a clearly defined centroid-matching evaluation protocol with bootstrap CIs and paired statistical tests; (2) fold-wise synthesis that prevents test leakage, with priors estimated only from training subjects; (3) transparent separation of reimplemented (R) and pretrained (L) comparison methods; (4) honest reporting of ablation results including non-significant p-values; (5) external validation on a different sequence (GRE→SWI) and scanner. The physics-guided synthesis module is a reasonable approach to the hard-negative scarcity problem, and the magnitude-only static-dephasing limitation is openly acknowledged. The patient-level burden analysis (Table II) is a valuable addition that honestly delineates the boundary between candidate extraction and calibrated burden estimation.

major comments (1)
  1. §IV-A, Table I caption: The VALDO 5-fold CV uses a single random seed for partition assignment (n=72, ~14 subjects per test fold). The headline F1 gain (74.3% vs 65.8% backbone, p=0.020) has substantially overlapping bootstrap CIs (±8.8% vs ±10.2%), and no individual ablation component reaches p<0.05 (Table III: centre-map p=0.0986; Table IV: all synthesis comparisons p>0.05). Without multi-seed CV or an independent held-out test set, the stability of the VALDO F1 improvement across partition choices is untested. The AIBL external result (n=370, p=0.0016) partially mitigates this, but AIBL is evaluated as the mean over five fold checkpoints (Table I caption), which is an ensemble-style evaluation that may overstate single-model performance. The authors should either (a) run multi-seed CV on VALDO and report the mean and standard deviation of the F1 metric across seeds, or (b) clearly add
minor comments (7)
  1. §III-C, Eq. (2): The focal centre-map loss Lctr uses the notation (Cx + epsilon_c)^gamma_c, but it is unclear whether this exponent applies to the target Cx or to the full expression. Clarifying the grouping would help readers.
  2. §III-D: The field-strength and echo-time ranges (1.5–3.0 T, 20–40 ms) are sampled when metadata is unavailable, but VALDO metadata includes scanner field strengths (1.5 T and 3 T). It would help to state how often metadata was actually unavailable and whether the sampled ranges match the VALDO acquisition distribution.
  3. Table I: The asterisk marking convention (p<0.05 against best local comparator) is defined in the caption but the specific comparator for each asterisked metric is not named. Adding the comparator name or a footnote would improve clarity.
  4. Table II: CenSynCMB's VALDO MAE (1.81) is bolded as best, but the FRST-style detector achieves 2.00 with a much higher Pearson r (0.84 vs 0.77). The caption should clarify whether bolding is per-metric or overall, and the text should note this tradeoff.
  5. §V-B: The Grad-CAM visualisation (Fig. 4) is discussed qualitatively but no quantitative evidence links the centre-map head to improved localisation beyond the ablation in Table III. A brief statement acknowledging this as qualitative evidence would be appropriate.
  6. Reference [23] is a medRxiv preprint (2026); if a peer-reviewed version is available, it should be updated.
  7. §I-C: The contributions list combines multiple techniques (centre-map, focal regression, uncertainty weighting, FN-driven reweighting) into a single bullet point. Separating the detector design contributions from the synthesis contribution would improve readability.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the careful reading and constructive statistical critique. The concern about single-seed partition stability on the small VALDO cohort is well-taken, and we will address it through both additional experiments and clearer reporting.

read point-by-point responses

  1. Referee: VALDO 5-fold CV uses a single random seed for partition assignment (n=72, ~14 subjects per test fold). Headline F1 gain has substantially overlapping bootstrap CIs, and no individual ablation component reaches p<0.05. Without multi-seed CV or an independent held-out test set, the stability of the VALDO F1 improvement across partition choices is untested.

    Authors: The referee is correct that the current VALDO evaluation relies on a single partition seed, and that the overlapping bootstrap CIs and non-significant ablation p-values mean the VALDO improvement cannot be considered definitively stable across partition choices. We acknowledge this limitation honestly. We will run multi-seed 5-fold cross-validation on VALDO (at least 5 distinct partition seeds) and report the mean and standard deviation of F1 (and other lesion-level metrics) across seeds. This will directly test whether the F1 improvement over the backbone and over local comparators is stable across partition choices. We will also report the proportion of seeds in which CenSynCMB outperforms the best local comparator, providing an intuitive stability measure. If the multi-seed results show that the improvement is not robust, we will revise the headline claim accordingly and reframe the VALDO result as trend-level rather than confirmatory. We agree this is necessary before the community can rely on the reported F1 gain. revision: yes

  2. Referee: AIBL is evaluated as the mean over five fold checkpoints (Table I caption), which is an ensemble-style evaluation that may overstate single-model performance.

    Authors: This is a fair and important observation. The current AIBL protocol averages predictions across five fold checkpoints, which is indeed an ensemble-style evaluation and does not reflect single-model deployment performance. We will address this in two ways. First, we will add a single-model AIBL evaluation: for each fold checkpoint, we will report AIBL metrics independently, and then report the mean and standard deviation across the five individual checkpoints. This will show the spread of single-model performance on AIBL. Second, we will clearly label the ensemble-style result as such in the table and text, so readers understand the distinction. We note that even under single-model evaluation, the AIBL cohort (n=370) is substantially larger than VALDO (n=72), so the external validation still provides meaningful evidence of cross-sequence transfer; however, the referee is right that the current presentation conflates single-model and ensemble performance, and we will correct this. revision: yes

  3. Referee: The authors should either (a) run multi-seed CV on VALDO and report the mean and standard deviation of the F1 metric across seeds, or (b) clearly add [comment appears truncated].

    Authors: We interpret option (b) as a request to clearly state the partition-stability limitation in the manuscript if multi-seed experiments are not feasible. We will in fact pursue option (a) as described above. Additionally, regardless of the multi-seed outcome, we will add an explicit limitation paragraph noting that VALDO is small (n=72), that the single-seed partition was the original design, and that ablation-level p-values did not reach conventional significance thresholds. We believe both actions together fully address the referee's concern. revision: yes

Circularity Check

0 steps flagged

No significant circularity found; the paper is an empirical detection study evaluated on external benchmarks with proper leakage prevention.

full rationale

The paper's derivation chain is straightforwardly empirical and does not contain any step where a prediction reduces to its inputs by construction. (1) The centre-map supervision (Eq. 1, §III-B) defines a Gaussian target around reference lesion centroids as an auxiliary training signal; it is not claimed as a first-principles derivation and is explicitly not used at inference time. (2) The training objective (Eqs. 2–3, §III-C) combines standard techniques (Tversky loss, focal regression, homoscedastic uncertainty weighting from Kendall et al. [11], FN-driven reweighting) — none of which are presented as novel derivations equivalent to their inputs. (3) The physics-guided synthesis (Eq. 4, §III-D) uses a standard dipole kernel cited from Salomir et al. [8], an independent external reference, and all synthesis priors are estimated only from training-fold subjects, with validation/test subjects explicitly excluded. (4) Evaluation uses VALDO Task 2 (5-fold CV) and AIBL (external, never used for training or model selection), with comparison methods either reimplemented under identical fold-wise protocol (R) or clearly separated as pretrained references (L). SHIVA-CMB v2, which used AIBL in development, is explicitly excluded from ranking. (5) The two self-citations ([24], [25]) appear only in the related work section for lacune/EPVS detection context and are not load-bearing for any method component or central claim. The ablation results are honestly reported as trend-level (p>0.05 for synthesis components). No fitted parameter is renamed as a prediction, no uniqueness theorem is invoked, and no ansatz is smuggled through self-citation. The minor self-citations in related work are non-load-bearing, warranting a score of 1 rather than 0.

Axiom & Free-Parameter Ledger

12 free parameters · 4 axioms · 2 invented entities

The paper introduces 12+ free parameters (loss weights, thresholds, synthesis ranges) that are set by the authors without derivation from first principles. The core domain assumption—that a static-dephasing dipole-kernel approximation suffices for training augmentation—is reasonable but unvalidated beyond trend-level ablation results. The two invented entities (synthetic mimics) lack independent evidence of realism.

free parameters (12)
  • alpha (Tversky FP weight) = 0.1
    Set to penalise false positives less than false negatives; chosen by the authors, not derived.
  • beta (Tversky FN weight) = 0.9
    Set to penalise false negatives more heavily; chosen by the authors.
  • sigma_c (centre-map Gaussian width) = 2 voxels
    Controls spatial spread of centre-map target; set by the authors.
  • gamma_c (focal centre-map exponent) = 2.5
    Focal parameter for centre-map regression; set by the authors.
  • gamma_s (crop size exponent) = 0.5
    Controls lesion-size reweighting; set by the authors.
  • rho (FN-response scaling) = 1.5
    Scales the false-negative-driven crop weight; set by the authors.
  • w_min, w_max (crop weight clip range) = [0.5, 5.0]
    Bounds on crop reweighting; set by the authors.
  • Probability threshold = 0.50
    Fixed post-processing threshold for detection; stated as fixed for all evaluations.
  • Minimum component size = 5 voxels
    Post-processing parameter for connected-component filtering.
  • chi (susceptibility amplitude) = Not specified
    Scalar susceptibility source amplitude for synthesis; value not given in the paper.
  • Field strength range = 1.5-3.0 T
    Domain randomization range for synthesis when metadata unavailable.
  • Echo time range = 20-40 ms
    Domain randomization range for synthesis when metadata unavailable.
axioms (4)
  • domain assumption CMBs are adequately represented as compact susceptibility sources whose local blooming pattern on T2*-GRE can be approximated by a static-dephasing magnitude model with a dipole kernel.
    Invoked in §III-D to justify the physics-guided synthesis. The paper acknowledges this is an approximation, not a full GRE/QSM simulator.
  • domain assumption The 5 mm centroid-matching criterion from VALDO Task 2 is an appropriate evaluation metric for CMB detection.
    Used throughout §IV-C for lesion-level evaluation. Inherited from the VALDO challenge protocol [1].
  • domain assumption Fold-wise estimation of lesion priors (counts, locations, sizes, intensity changes) from training subjects generalises to held-out subjects.
    Underlies the synthesis module design in §III-D. Not independently tested.
  • ad hoc to paper Magnitude-only static dephasing is sufficient for training augmentation despite inability to distinguish calcification from blood by susceptibility sign.
    Stated in §III-D and §VI-C. The paper acknowledges this as a limitation but uses it anyway for hard-negative synthesis.
invented entities (2)
  • Synthetic calcification-like hard negatives no independent evidence
    purpose: Train detector to reject compact low-signal non-CMB structures
    Constructed from compact low-signal sources with non-CMB multi-contrast patterns; the specific multi-contrast pattern definition is underspecified and no external validation of realism is provided beyond Fig. 3 visual comparison.
  • Synthetic vessel-like hard negatives no independent evidence
    purpose: Train detector to reject tubular low-signal structures
    Constructed from elongated anisotropic masks; no quantitative validation that synthesised vessels match real vessel false-positive distributions.

pith-pipeline@v1.1.0-glm · 16234 in / 3274 out tokens · 184963 ms · 2026-07-07T17:05:00.303132+00:00 · methodology

0 comments
read the original abstract

Cerebral microbleeds (CMBs) are MRI markers of small vessel disease and the microbleed component of amyloid related imaging abnormalities (ARIA-H), but their small size, sparsity, and similarity to vessels, calcification-like foci, and artefacts make automated detection difficult. We propose CenSynCMB, a centre-guided and mimic-aware framework combining a 3D Attention U-Net, auxiliary centre-map supervision, false-negative-driven reweighting, and fold-wise physics-guided synthesis of positive CMBs and labelled hard negatives. Synthetic data expose the detector to compact lesions and common mimics without validation or test leakage. On VALDO Task 2, CenSynCMB achieved the best local-comparison lesion-level F1 (74.3%, p = 0.020); on external AIBL SWI, it achieved the highest local-comparison recall (88.5%, p = 0.0058) and F1 (65.0%, p = 0.0016). Together, these results support scalable CMB candidate extraction in large, unlabelled MRI cohorts, while highlighting cohort-specific calibration as the next step toward reliable burden estimation.

Figures

Figures reproduced from arXiv: 2607.05325 by Adama Fatima Saccoh, Carole H. Sudre, Frederik Barkhof, Hanyuan Zhang, Krinos Li, Lucas He, Marleen de Bruijne, Rafael Rehwald, Rhodri Davies, Silvia Ingala.

Figure 1
Figure 1. Figure 1: Representative appearance of cerebral microbleeds (CMBs) on T2*-weighted gradient-echo MRI (T2*-GRE, VALDO), shown for two subjects. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the centre-guided, synthesis-augmented CMB detection framework. (A) Real VALDO data provide T2S-space co-registered T2 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative examples of real and synthesized CMB-related appear [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Grad-CAM visualisation of the centre-guided detector outputs, [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

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

25 extracted references · 25 canonical work pages · 4 internal anchors

  1. [1]

    Where is V ALDO? V Ascular Lesions Detection and segmentatiOn challenge at MICCAI 2021,

    C. H. Sudre et al., “Where is V ALDO? V Ascular Lesions Detection and segmentatiOn challenge at MICCAI 2021,” Medical Image Analysis, 2024

    work page 2021

  2. [2]

    Neuroimaging standards for research into small vessel disease and its contribution to ageing and neurodegeneration,

    J. M. Wardlaw et al., “Neuroimaging standards for research into small vessel disease and its contribution to ageing and neurodegeneration,” Lancet Neurology, vol. 12, no. 8, pp. 822–838, 2013

    work page 2013

  3. [3]

    Cerebral microbleeds: a guide to detection and interpretation,

    S. M. Greenberg et al., “Cerebral microbleeds: a guide to detection and interpretation,” Lancet Neurology, vol. 8, no. 2, pp. 165–174, 2009

    work page 2009

  4. [4]

    The Microbleed Anatomical Rating Scale (MARS): reliability of a tool to map brain microbleeds,

    S. M. Gregoire et al., “The Microbleed Anatomical Rating Scale (MARS): reliability of a tool to map brain microbleeds,” Neurology, vol. 73, no. 21, pp. 1759–1766, 2009

    work page 2009

  5. [5]

    Improving interrater agreement about brain microbleeds: development of the Brain Observer MicroBleed Scale (BOMBS),

    C. Cordonnier et al., “Improving interrater agreement about brain microbleeds: development of the Brain Observer MicroBleed Scale (BOMBS),” Stroke, vol. 40, no. 1, pp. 94–99, 2009

    work page 2009

  6. [6]

    S. Y . Jeong et al., “Amyloid-related imaging abnormalities in the era of anti-amyloid beta monoclonal antibodies for Alzheimer’s disease: recent updates on clinical and imaging features and MRI monitoring,” Korean Journal of Radiology, vol. 25, no. 8, pp. 726–741, 2024

    work page 2024

  7. [7]

    Synthetic Cerebral Microbleed on SWI Images,

    S. Momeni et al., “Synthetic Cerebral Microbleed on SWI Images,” CSIRO Data Access Portal, Collection 50304, 2021, doi: 10.25919/aegy- ny12

    work page doi:10.25919/aegy- 2021

  8. [8]

    A fast calculation method for magnetic field inhomogeneity due to an arbitrary distribution of bulk susceptibility,

    R. Salomir, B. D. de Senneville, and C. T. W. Moonen, “A fast calculation method for magnetic field inhomogeneity due to an arbitrary distribution of bulk susceptibility,” Concepts in Magnetic Resonance Part B: Magnetic Resonance Engineering, vol. 19B, no. 1, pp. 26–34, 2003

    work page 2003

  9. [9]

    MONAI: An open-source framework for deep learning in healthcare

    M. J. Cardoso et al., “MONAI: an open-source framework for deep learning in healthcare,” arXiv:2211.02701, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  10. [10]

    Isambard-AI: a leadership class supercomputer optimised specifically for Artificial Intelligence

    S. McIntosh-Smith, S. R. Alam, and C. Woods, “Isambard-AI: a leadership class supercomputer optimised specifically for artificial in- telligence,” arXiv:2410.11199, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  11. [11]

    Multi-task learning using uncer- tainty to weigh losses for scene geometry and semantics,

    A. Kendall, Y . Gal, and R. Cipolla, “Multi-task learning using uncer- tainty to weigh losses for scene geometry and semantics,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), pp. 7482–7491, 2018

    work page 2018

  12. [12]

    nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation,

    F. Isensee, P. F. Jaeger, S. A. A. Kohl, J. Petersen, and K. H. Maier-Hein, “nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation,” Nature Methods, vol. 18, no. 2, pp. 203–211, 2021

    work page 2021

  13. [13]

    Attention U-Net: Learning Where to Look for the Pancreas

    O. Oktay et al., “Attention U-Net: Learning where to look for the pancreas,” arXiv:1804.03999, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  14. [14]

    Swin UNETR: Swin transformers for semantic segmentation of brain tumors in MRI images,

    A. Hatamizadeh, V . Nath, Y . Tang, D. Yang, H. R. Roth, and D. Xu, “Swin UNETR: Swin transformers for semantic segmentation of brain tumors in MRI images,” in Brainlesion (BrainLes 2021), LNCS 12962, pp. 272–284, 2022

    work page 2021

  15. [15]

    VISTA3D: A unified segmentation foundation model for 3D medical imaging,

    Y . He et al., “VISTA3D: A unified segmentation foundation model for 3D medical imaging,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2025

    work page 2025

  16. [16]

    A multi-stage framework for cerebral microbleeds detection and segmentation,

    Z. Chen and Q. Dou, “A multi-stage framework for cerebral microbleeds detection and segmentation,” MICCAI 2021 V ALDO Challenge (Task 2), 2021

    work page 2021

  17. [17]

    MixMicrobleedNet: segmentation of cerebral microbleeds using nnU-Net

    H. J. Kuijf, “MixMicrobleedNet: segmentation of cerebral microbleeds using nnU-Net,” arXiv:2108.01389, 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  18. [18]

    Automated detection of cerebral microbleeds in MR images: A two- stage deep learning approach,

    M. A. Al-masni, W.-R. Kim, E. Y . Kim, Y . Noh, and D.-H. Kim, “Automated detection of cerebral microbleeds in MR images: A two- stage deep learning approach,” NeuroImage: Clinical, vol. 28, p. 102464, 2020

    work page 2020

  19. [19]

    Toward automated detection of microbleeds with anatomical scale localization using deep learning,

    J.-H. Kim et al., “Toward automated detection of microbleeds with anatomical scale localization using deep learning,” Medical Image Analysis, vol. 101, p. 103415, 2025

    work page 2025

  20. [20]

    Automated detection of cerebral microbleeds on MR images using knowledge distillation framework,

    V . Sundaresan et al., “Automated detection of cerebral microbleeds on MR images using knowledge distillation framework,” Frontiers in Neuroinformatics, vol. 17, 2023

    work page 2023

  21. [21]

    SHIV A-CMB: a deep-learning-based robust cerebral microbleed segmentation tool trained on multi-source T2*GRE- and susceptibility-weighted MRI,

    A. Tsuchida et al., “SHIV A-CMB: a deep-learning-based robust cerebral microbleed segmentation tool trained on multi-source T2*GRE- and susceptibility-weighted MRI,” Scientific Reports, vol. 14, art. 30901, 2024

    work page 2024

  22. [22]

    A robust deep learning framework for cerebral microbleeds recognition in GRE and SWI MRI,

    T. Hassanzadeh, S. Sachdev, W. Wen, P. S. Sachdev, and A. Sowmya, “A robust deep learning framework for cerebral microbleeds recognition in GRE and SWI MRI,” NeuroImage: Clinical, p. 103873, 2025

    work page 2025

  23. [23]

    Automated quantification of cerebral mi- crobleeds for ARIA-H monitoring in aging and Alzheimer’s dis- ease: a multicenter deep learning validation,

    Z. X. B. Low et al., “Automated quantification of cerebral mi- crobleeds for ARIA-H monitoring in aging and Alzheimer’s dis- ease: a multicenter deep learning validation,” medRxiv, 2026, doi: 10.64898/2026.05.19.26353364

    work page doi:10.64898/2026.05.19.26353364 2026

  24. [24]

    Compressing kv cache for long-context llm inference with inter-layer attention similarity

    L. He et al., “SWIN-DS: A deeply supervised transformer with geometric guidance for robust lacune detection,” in Proc. IEEE Int. Conf. Acoust., Speech Signal Process. (ICASSP), 2026, doi: 10.1109/ICASSP55912.2026.11461885

    work page doi:10.1109/icassp55912.2026.11461885 2026

  25. [25]

    A unified framework for joint detection of lacunes and enlarged perivascular spaces,

    L. He et al., “A unified framework for joint detection of lacunes and enlarged perivascular spaces,” arXiv:2603.04243, 2026

    work page arXiv 2026