Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Detection of Breast Cancer Lumpectomy Margin with SAM-incorporated Forward-Forward Contrastive Learning

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a self-supervised patch classifier followed by prompt-based SAM refinement detects and delineates positive lumpectomy margins on specimen radiographs with AUC 0.8455, a 27.4% Dice gain over the coarse-mask baseline…

desk verdict The pipeline is real and the engineering is sound, but the evaluation only includes positive-margin patients, so the clinical claim in the abstract outruns the data. read the letter →

arxiv 2506.21006 v1 pith:DXXWLLAF submitted 2025-06-26 cs.CV

Tyler Ward , Xiaoqin Wang , Braxton McFarland , Md Atik Ahamed , Sahar Nozad , Talal Arshad , Hafsa Nebbache , Jin Chen
show 1 more author
Abdullah Imran
This is my paper
classification cs.CV
keywords breastcancerlumpectomymarginspecimenradiographyforward-forwardalgorithmcontrastivelearningSegmentAnythingModelmedicalimagesegmentationintraoperativeassessment
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that combining forward-forward contrastive learning (FFCL) with Segment Anything Model (SAM) refinement can detect and delineate positive lumpectomy margins on specimen radiographs accurately enough to be useful in the operating room. It reports an AUC of 0.8455 for patch-level margin classification, a 27.4% improvement in Dice similarity over the coarse-mask baseline, and 47 ms inference per image. If correct, this would address a real clinical gap: conventional specimen radiography misses many positive margins, and roughly one in four lumpectomy patients currently needs a second surgery. The practical payoff would be an intraoperative margin map that lets the surgeon re-excise a positive margin during the same operation.

What carries the argument

The load-bearing mechanism is the two-stage FFCL-SAM pipeline. FFCL is a backpropagation-free pretraining strategy that applies local contrastive learning at each layer (the forward-forward part) and then global contrastive learning across the network, producing patch embeddings without needing labeled data. A ResNet-18 is fine-tuned with focal loss to classify $64 \times 64$ patches, and those patch labels are remapped to a coarse binary mask. The refinement stage uses SAM 2 prompted by two things generated from that coarse mask: a bounding box that encloses it and the coarse mask itself as a mask prompt. The bounding box produces an initial mask, and the mask prompt refines it into the final segmentation; this two-prompt design is what lets a generic foundation model correct the blocky artifacts of patch-based classification.

What would settle it

Run FFCL-SAM on a held-out cohort that includes both pathology-confirmed positive-margin and negative-margin lumpectomies, convert the predicted margin masks into a patient-level binary call (any predicted positive margin), and compare sensitivity and specificity against pathology. If sensitivity on that mixed cohort is not better than the 36–58% range reported for conventional specimen radiography, or if specificity is too low to be clinically acceptable, the clinical claim fails.

Watch

Extended reading notes

Core claim

The central claim is that FFCL pretraining followed by SAM-based mask refinement turns specimen radiographs into reliable margin segmentations without manual prompting. A ResNet-18 backbone is pretrained with two contrastive stages—local forward-forward learning at each layer and global contrastive learning across the network—then fine-tuned with focal loss to classify $64 \times 64$ patches as positive or negative margin. Reconstructing those patch labels into a coarse binary mask and prompting SAM 2 with a bounding box plus the mask itself yields the final refined mask. In the paper's experiments, this pipeline reaches AUC 0.8455 for margin classification, improves positive-margin Dice from 0.6313 (coarse reconstructed mask) to 0.8861 with zero-shot SAM 2 and 0.9053 with five-image few-shot fine-tuning, and runs mask refinement in 47 ms per image. The validation and test sets deliberately contain only positive-margin patients, so the reported numbers measure how well the model detects margins when they are present.

Load-bearing premise

The load-bearing premise is that patch-level accuracy measured only on positive-margin patients transfers to patient-level decisions in a real cohort that also contains negative-margin patients.

Editorial extensions

If this is right

  • If FFCL-SAM works as reported, a surgeon could receive a refined margin map during the operation, in under 50 ms per image, and re-excise a positive margin immediately instead of sending the patient for a second surgery.
  • The two-prompt refinement removes the manual prompting bottleneck of SAM, because the coarse mask itself supplies both the bounding box and the mask prompt, so the foundation model runs automatically.
  • The backpropagation-free FFCL pretraining stage makes the approach viable in compute-constrained settings: the ResNet-18 variant pretrains in about 50 minutes in the paper's setup, far faster than the ViT alternative.
  • The modular design implies the SAM refinement step can be attached to any patch-level margin classifier, so improvements in the coarse mask and improvements in the refinement are separable and independently testable.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper does not test this, but the 27.4% Dice gain over the coarse mask could partly reflect SAM smoothing the blocky patch grid rather than detecting new margin tissue; comparing SAM refinement on true versus randomly shifted coarse masks would separate those effects.
  • Because the validation and test sets contain only positive-margin patients, the AUC 0.8455 is a detection-among-positives number; the clinically relevant test is a mixed cohort of positive and negative margins, where patient-level sensitivity and specificity can be measured.
  • The framework is not tied to specimen radiography; the same FFCL patch classifier plus SAM refinement could be pointed at other intraoperative margin modalities, such as ultrasound or fluorescence imaging, given comparable patch-level labels.
  • If patient-level sensitivity holds up in a mixed cohort, the method could reduce re-excision rates by localizing the re-excision to the positive region, but that conclusion requires prospective validation, not just patch metrics.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes FFCL-SAM, a two-stage pipeline for lumpectomy margin detection on intraoperative specimen radiographs. Stage one uses Forward-Forward Contrastive Learning (FFCL) to pre-train a ResNet-18 (or ViT/ConvNeXt) backbone for patch-level classification of positive versus negative margin tissue; the patch predictions are reconstructed into a coarse binary mask. Stage two uses the coarse mask to prompt SAM or SAM 2 for refined segmentation. Experiments on a private dataset of 46 patients report patch-level classification AUC up to 0.8455 and segmentation Dice up to 0.9248 with SAM2-few-shot, plus a 47 ms inference time. The abstract claims that the method significantly enhances the speed and accuracy of intraoperative margin assessment and has the potential to reduce re-excision rates.

Significance. If the evaluation were valid, the integration of a backpropagation-free contrastive pre-training method with a promptable foundation model would be a practically interesting contribution to intraoperative margin assessment. The paper also provides a public code repository, uses a real annotated clinical dataset, and includes ablations over architecture and loss settings. However, the current experimental design does not support the central clinical claims: the test set contains only positive-margin patients, classification metrics are computed at the patch level, and the few-shot SAM2 protocol is not described with enough detail to rule out leakage. These are load-bearing issues because the claimed reduction in re-excision rates depends on specificity and patient-level margin classification, neither of which is measured.

major comments (4)
  1. [§4.1, Table 1] The validation and test sets contain only positive-margin patients (Table 1: 'Only positive cases are used for validation and testing'), so the classifier is never evaluated on images from negative-margin patients. All classification metrics in Table 2 are therefore patch-level measures on positive patients only, and specificity for the clinically relevant patient-level decision is never measured. The abstract's claim that FFCL-SAM 'enhances the speed and accuracy of intraoperative margin assessment' and can 'reduce re-excision rates' is not supported by this protocol, because false positives on normal tissue and correct identification of margin-negative specimens are untested.
  2. [§4.2, Table 3] The SAM2-few-shot experiments fine-tune the model on 'just five labeled SRs' but the text never states where these five images come from (e.g., training split, validation split, or held-out test images) or how they were selected. If any of the five images are from the test set, the reported DSC values in Table 3 are leakage-inflated. The source and selection procedure for these images must be reported, and the experiment must be rerun if any test image was used.
  3. [Abstract and §4.4] The headline '27.4% improvement in Dice similarity over baseline models' does not match Table 3. The overall DSC goes from 0.6587 (reconstructed mask) to 0.9248 (SAM2-few-shot), a relative improvement of about 40%, or an absolute increase of 0.266 (26.6 percentage points). The text's statement that SAM2-zero-shot boosts DSC 'by 25.48%' is also an absolute percentage-point difference (0.8861 − 0.6313), not a relative improvement. Please report improvements consistently as relative or absolute changes and correct the abstract accordingly.
  4. [§4.3, Table 2] All classification performance metrics are computed at the patch level (64×64 patches). Since patches from the same patient are highly correlated, standard deviations over 5 runs do not capture clustering by patient, and the test set contains only 6 positive patients. To support the margin-assessment claim, the authors should report image-level or specimen-level classification statistics (e.g., whether at least one positive patch is detected per specimen) and account for within-patient correlation.
minor comments (6)
  1. [§4.3, Eq. (10)] The formula for AUC is not the standard ROC definition; it should be written as an integral over the false-positive rate of the true-positive rate at the corresponding threshold.
  2. [§4.3, Eq. (9)] The F1 formula contains a stray period before the denominator ('... ×Precision +Recall'); please correct the typo.
  3. [§3.3.4] The sentence 'This mask provides an initial visualization of a potential positive margin but does not lack spatial precision' contains a double negative; it should read 'lacks spatial precision.'
  4. [§4.2] The machine configuration lists 'dual NVIDIA307 A4000X2 GPUs,' which appears to be a typo for 'NVIDIA RTX A4000 X2 GPUs.'
  5. [Table 3] The meaning of 'Negative Margin' in Table 3 is unclear, since Section 4.1 states that patches are extracted from positive margin regions only for validation and testing; please clarify what the negative margin segmentation refers to and how its ground truth was obtained.
  6. [References] Reference [26] lists the first author as 'Hiagen Hu'; this is likely a typo for 'Haigen Hu.'

Circularity Check

1 steps flagged · score 2.0 of 10

Minor self-citation for FFCL's claimed superiority; headline AUC/Dice results are empirical and not circular.

  1. self citation load bearing [Section 4.4, Results and Discussion, paragraph 'Margin detection via classification (FFCL)']
    "FFCL demonstrates the effectiveness in detecting lumpectomy margins from SRs, outperforming its non-pretraining counterparts with the same architecture (e.g., ResNet-18 w/ FFCL over ResNet-18 only) [2]."

    The paper attributes the central benefit of FFCL pretraining to reference [2], the authors' own IWBI 2024 paper, rather than to any comparison included in the present manuscript. Table 2 only compares three FFCL-pretrained backbones (ResNet-18, ViT, ConvNeXt) against each other; it does not compare FFCL pretraining against no pretraining or against other contrastive strategies. Reference [1], also by the same group, is likewise invoked for the assertion that FFCL beats other contrastive learning strategies. Thus the premise that FFCL is responsible for the accuracy improvement rests on a self-citation chain rather than on an independent comparison in this paper.

full rationale

The paper is primarily an empirical evaluation: patches are extracted from annotated specimen radiographs, an FFCL-pretrained backbone is fine-tuned for patch classification, coarse masks are reconstructed from patch labels, and SAM/SAM2 refine those masks. The headline metrics (AUC, DSC, HD, accuracy, inference time) are measured against held-out annotations and do not reduce by construction to fitted parameters or to prior equations. I found no instance where a predicted quantity is defined in terms of the target quantity, and no fitted value is renamed as a prediction. The evaluation protocol's use of only positive-margin patients in validation and testing is a real external-validity concern for the claimed clinical benefit, but it is not a circularity under the stated rules, so it does not raise the score. The only circularity-adjacent element is the self-cited support for FFCL's superiority over other pretraining strategies: the paper cites the authors' own prior works [1,2] for this claim instead of demonstrating it with an in-paper baseline. Because the central measured results are independent of those citations, the appropriate score is low.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim depends on a set of domain assumptions about the accuracy of annotations and the transferability of patch-level tumor classification to margin detection in a mixed patient population. The paper also assumes that AUC > 0.8 is clinically useful based on a cited reference. These are reasonable but untested assumptions in the presented evaluation.

free parameters (6)
  • focal_loss_alpha = 0.8
    Tuned on the validation set via ROC comparison (Fig. 6).
  • focal_loss_gamma = 3.0
    Tuned on the validation set via ROC comparison (Fig. 6).
  • patch_size = 64 x 64
    Chosen for patch extraction without a stated optimization criterion (Section 3.3.2).
  • positive_patch_stride = 3
    Set to 3 to oversample the minority positive class (Section 3.3.2).
  • negative_patch_stride = 45 (train), 35 (validation), 60 (test)
    Hand-selected per split; no criterion is provided (Section 3.3.2).
  • sam2_few_shot_fine_tuning_images = 5 labeled specimen radiographs
    Used to fine-tune SAM 2; the source of these images (train or test) is not stated (Section 4.4).
assumptions (4)
  • domain assumption Pathology-confirmed margin annotations made with ITK-SNAP are accurate ground truth.
    The experiments rely on human annotations as labels for training and evaluation (Section 4.1).
  • domain assumption Patches from the tumor center (positive) and non-malignant tissue (negative) are sufficient training data for detecting positive margins at test time.
    No positive margin patches are used in training; margin detection is framed as a transfer task (Section 4.1).
  • domain assumption AUC greater than 0.8 indicates potential clinical usefulness.
    The paper invokes this threshold from reference [13] to interpret the AUC (Section 4.3).
  • domain assumption SAM 2 zero-shot refinement improves coarse mask quality without task-specific training.
    The pipeline assumes the foundation model will refine the reconstructed mask, and reports improved Dice (Section 3.3.4).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Detection of Breast Cancer Lumpectomy Margin with SAM-incorporated Forward-Forward Contrastive Learning." pith.science (2026). https://pith.science/paper/DXXWLLAF

@misc{pith2026250621006,
  author       = {Pith},
  title        = {Pith review of: Detection of Breast Cancer Lumpectomy Margin with SAM-incorporated Forward-Forward Contrastive Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DXXWLLAF}},
  note         = {Machine review of arXiv:2506.21006}
}
read the original abstract

Complete removal of cancer tumors with a negative specimen margin during lumpectomy is essential in reducing breast cancer recurrence. However, 2D specimen radiography (SR), the current method used to assess intraoperative specimen margin status, has limited accuracy, resulting in nearly a quarter of patients requiring additional surgery. To address this, we propose a novel deep learning framework combining the Segment Anything Model (SAM) with Forward-Forward Contrastive Learning (FFCL), a pre-training strategy leveraging both local and global contrastive learning for patch-level classification of SR images. After annotating SR images with regions of known maligancy, non-malignant tissue, and pathology-confirmed margins, we pre-train a ResNet-18 backbone with FFCL to classify margin status, then reconstruct coarse binary masks to prompt SAM for refined tumor margin segmentation. Our approach achieved an AUC of 0.8455 for margin classification and segmented margins with a 27.4% improvement in Dice similarity over baseline models, while reducing inference time to 47 milliseconds per image. These results demonstrate that FFCL-SAM significantly enhances both the speed and accuracy of intraoperative margin assessment, with strong potential to reduce re-excision rates and improve surgical outcomes in breast cancer treatment. Our code is available at https://github.com/tbwa233/FFCL-SAM/.

Figures

Figures reproduced from arXiv: 2506.21006 by the authors.

Figure 1
Figure 1. Overview of the proposed FFCL-SAM workflow: First, FFCL is employed to predict margin status at the patch-level on an SR [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The proposed Forward-Forward Contrastive Learning (FFCL) for breast cancer margin detection. FFCL training involves [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Detailed illustration of the breast cancer margin detection pipeline consisting of both classification and segmentation tasks. [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Sample patch extraction process from positive and negative margin regions of two different SR images. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Anterior-posterior (AP), craniocaudal (CC), and mediolateral (MLO) image samples from our dataset with their corresponding [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: Visualization of Receiver operating curve (ROC) curves comparing the binary cross-entropy and Focal cross-entropy losses (w/ [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 7
Figure 7. Figure 7: Visual comparison of the segmented margins from different SR images. Note that in each case, the SAM 2 refinement process [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. A Probabilistic Segment Anything Model for Ambiguity-Aware Medical Image Segmentation

    cs.CV 2025-09 conditional novelty 4.0 of 10

    A SAM variant that samples a latent code into the prompt embedding produces diverse lung nodule masks and reports better GED, DSC, and IoU than Probabilistic U-Net on LIDC-IDRI.

Reference graph

Works this paper leans on

61 extracted references · 52 canonical work pages · cited by 1 Pith paper

  1. [1]

    Md Atik Ahamed, Jin Chen, and Abdullah-Al-Zubaer Imran. 2023. FFCL: Forward-Forward Contrastive Learning for Improved Medical Image Classification. In Medical Imaging with Deep Learning (MIDL), short paper track

    work page 2023

  2. [2]

    Atik Ahamed, Braxton McFarland, Xiaoqin Wang, Jin Chen, and Abdullah-Al-Zubaer Imran

    Md. Atik Ahamed, Braxton McFarland, Xiaoqin Wang, Jin Chen, and Abdullah-Al-Zubaer Imran. 2024. Automatic detection of breast cancer lumpectomy margin from intraoperative specimen mammography. In 17th International Workshop on Breast Imaging (IWBI 2024) , Vol. 13174. SPIE, 450–454. Detection of Breast Cancer Lumpectomy Margin with SAM-incorporated Forward...

    work page 2024

  3. [3]

    Sana Ahuja, Priya Yadav, Marzieh Fattahi-Darghlou, and Sufian Zaheer. 2024. Comparison of Intraoperative Imprint Cytology versus Frozen Section for Sentinel Lymph Node Evaluation in Breast Cancer. A study along with Systematic Review and Meta-analysis of literature. Asian Pacific Journal of Cancer Prevention: APJCP 25, 4 (2024)

    work page 2024

  4. [4]

    Muhammad Awais, Muzammal Naseer, Salman Khan, Rao Muhammad Anwer, Hisham Cholakkal, Mubarak Shah, Ming-Hsuan Yang, and Fa- had Shahbaz Khan. 2025. Foundation Models Defining a New Era in Vision: A Survey and Outlook. IEEE Transactions on Pattern Analysis and Machine Intelligence (2025)

    work page 2025

  5. [5]

    Reza Azad, Ehsan Khodapanah Aghdam, Amelie Rauland, Yiwei Jia, Atlas Haddadi Avval, and Afshin Bozorgpour. 2024. Medical Image Segmentation Review: The Success of U-Net. IEEE Transactions on Pattern Analysis and Machine Intelligence 46, 12 (2024), 10076–10095

    work page 2024

  6. [6]

    Shekoofeh Azizi, Basil Mustafa, Fiona Ryan, Zachary Beaver, Jan Freyberg, Jonathan Deaton, Aaron Loh, Alan Karthikesalingam, Simon Kornblith, Ting Chen, Vivek Natarajan, and Mohammad Norouzi. 2021. Big self-supervised models advance medical image classification. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) . 3478–3488

    work page 2021

  7. [7]

    Hu Cao, Yueyue Wang, Joy Chen, Dongsheng Jiang, Xiaopeng Zhang, Qi Tian, and Manning Wang. 2022. Swin-Unet: Unet-like pure transformer for medical image segmentation. In European Conference on Computer Vision (ECCV) . Springer, 205–218

    work page 2022

  8. [8]

    Jeffery M Chakedis, Annie Tang, Alison Savitz, Liisa L Lyon, Patricia E Palacios, Brooke Vuong, Maihgan A Kavanagh, Gillian E Kuehner, Sharon B Chang, and Permanente Medical Group Breast Research Collaborative. 2022. Economic impact of reducing reexcision rates after breast-conserving surgery in a large, integrated health system. Annals of Surgical Oncolo...

    work page 2022

Show all 61 references
  1. [9]

    Jieneng Chen, Yongyi Lu, Qihang Yu, Xiangde Luo, Ehsan Adeli, Yan Wang, Le Lu, Alan L Yuille, and Yuyin Zhou. 2021. TransUNet: Transformers make strong encoders for medical image segmentation. arXiv preprint arXiv:2102.04306 (2021)

    2021 arXiv

  2. [10]

    Jieneng Chen, Jieru Mei, Xianhang Li, Yongyi Lu, Qihang Yu, Qingyue Wei, Xiangde Luo, Yutong Xie, Ehsan Adeli, Yan Wang, et al. 2024. TransUNet: Rethinking the U-Net architecture design for medical image segmentation through the lens of transformers. Medical Image Analysis 97 (2024)

    2024

  3. [11]

    Kevin A Chen, Kathryn E Kirchoff, Logan R Butler, Alexa D Holloway, Muneera R Kapadia, Cherie M Kuzmiak, Stephanie M Downs-Canner, Phillip M Spanheimer, Kristalyn K Gallagher, and Shawn M Gomez. 2023. Analysis of Specimen Mammography with Artificial Intelligence to Predict Mar...

    2023

  4. [12]

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. A simple framework for contrastive learning of visual representations. In Proceedings of the 37th International Conference on Machine Learning , Vol. 119. PMLR, 1597–1607

    2020

  5. [13]

    Şeref Kerem Çorbacıoğlu and Gökhan Aksel. 2023. Receiver operating characteristic curve analysis in diagnostic accuracy studies: A guide to interpreting the area under the curve value. Turkish journal of emergency medicine 23, 4 (2023), 195–198

    2023

  6. [14]

    Sandryne David, Trang Tran, Frédérick Dallaire, Guillaume Sheehy, Feryel Azzi, Dominique Trudel, Francine Tremblay, Atilla Omeroglu, Frédéric Leblond, and Sarkis Meterissian. 2023. In situ Raman spectroscopy and machine learning unveil biomolecular alterations in invasive brea...

    2023

  7. [15]

    Alexey Dosovitskiy. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929 (2020)

    2020 arXiv

  8. [16]

    Gavin P Dowling, Cian M Hehir, Gordon R Daly, Sandra Hembrecht, Stephen Keelan, Katie Giblin, Maen M Alrawashdeh, Fiona Boland, and Arnold DK Hill. 2024. Diagnostic accuracy of intraoperative methods for margin assessment in breast cancer surgery: A systematic review & meta-an...

    2024

  9. [17]

    Timothy M D’Alfonso, David Joon Ho, Matthew G Hanna, Anne Grabenstetter, Dig Vijay Kumar Yarlagadda, Luke Geneslaw, Peter Ntiamoah, Thomas J Fuchs, and Lee K Tan. 2021. Multi-magnification-based machine learning as an ancillary tool for the pathologic assessment of shaved marg...

    2021

  10. [18]

    Hospedales

    Linus Ericsson, Henry Gouk, Chen Change Loy, and Timothy M. Hospedales. 2022. Self-supervised representation learning: Introduction, advances, and challenges. IEEE Signal Processing Magazine 39, 3 (2022), 42–62

    2022

  11. [19]

    Karla Esbona, Zhanhai Li, and Lee G Wilke. 2012. Intraoperative imprint cytology and frozen section pathology for margin assessment in breast conservation surgery: a systematic review. Annals of Surgical Oncology 19 (2012), 3236–3245

    2012

  12. [20]

    Kyana Y Garza, Mary E King, Chandandeep Nagi, Rachel J DeHoog, Jialing Zhang, Marta Sans, Anna Krieger, Clara L Feider, Alena V Bensussan, Michael F Keating, et al. 2024. Intraoperative Evaluation of Breast Tissues During Breast Cancer Operations Using the MasSpec Pen. JAMA Ne...

    2024

  13. [21]

    Jean-Bastien Grill, Florian Strub, Florent Altché, Corentin Tallec, Pierre Richemond, Elena Buchatskaya, Carl Doersch, Bernardo Avila Pires, Zhaohan Guo, Mohammad Gheshlaghi Azar, Bilal Piot, koray kavukcuoglu, Remi Munos, and Michal Valko. 2020. Bootstrap Your Own Latent - A ...

    2020

  14. [22]

    Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. 2020. Momentum Contrast for unsupervised visual representation learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 9729–9738

    2020

  15. [23]

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep Residual Learning for Image Recognition. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 770–778

    2016

  16. [24]

    Geoffrey Hinton. 2022. The Forward-Forward Algorithm: Some preliminary investigations. arXiv preprint arXiv:2212.13345 (2022)

    2022 arXiv

  17. [25]

    David Joon Ho, Narasimhan P Agaram, Peter J Schüffler, Chad M Vanderbilt, Marc-Henri Jean, Meera R Hameed, and Thomas J Fuchs. 2020. Deep interactive learning: An efficient labeling approach for deep learning-based osteosarcoma treatment response assessment. In Medical Image C...

    2020

  18. [26]

    Hiagen Hu, Xiaoyuan Wang, Yan Zhang, Qi Chen, and Qiu Guan. 2024. A comprehensive survey on contrastive learning. Neurocomputing 610 (2024)

    2024

  19. [27]

    Weijian Huang, Cheng Li, Hong-Yu Zhou, Hao Yang, Jiarun Liu, Yong Liang, Hairong Zheng, Shaoting Zhang, and Shanshan Wang. 2024. Enhancing representation in radiography-reports foundation model: a granular alignment algorithm using masked contrastive learning.Nature Communicat...

    2024

  20. [28]

    Rajini Katipamula, Amy C Degnim, Tanya Hoskin, Judy C Boughey, Charles Loprinzi, Clive S Grant, Kathleen R Brandt, Sandhya Pruthi, Christopher G Chute, Janet E Olson, et al. 2009. Trends in mastectomy rates at the Mayo Clinic Rochester: Effect of surgical year and preoperative...

    2009

  21. [29]

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C Berg, Wan-Yen Lo, et al. 2023. Segment anything. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) . 4015–4026

    2023

  22. [30]

    Meeghan Lautner, Heather Lin, Yu Shen, Catherine Parker, Henry Kuerer, Simona Shaitelman, Gildy Babiera, and Isabelle Bedrosian. 2015. Disparities in the use of breast-conserving therapy among patients with early-stage breast cancer. JAMA Surgery 150, 8 (2015), 778–786

    2015

  23. [31]

    Le-Khac, Graham Healy, and Alan F

    Phuc H. Le-Khac, Graham Healy, and Alan F. Smeaton. 2020. Contrastive representation learning: A framework and review. IEEE Access 8 (2020), 193907–193934

    2020

  24. [32]

    Pengcheng Li, Li Zhihao, Zijian Wang, Li Chaoxiang, and Monan Wang. 2024. mResU-Net: multi-scale residual U-Net-based brain tumor segmentation from multimodal MRI. Medical & Biological Engineering & Computing 62, 3 (2024), 641–651

    2024

  25. [33]

    Zheren Li, Zhiming Cui, Sheng Wang, Yuji Qi, Xi Ouyang, Qitian Chen, Yuezhi Yang, Zhong Xue, Dinggang Shen, and Jie-Zhi Cheng. 2021. Domain generalization for mammography detection via multi-style and multi-view contrastive learning. In Medical Image Computing and Computer Ass...

    2021

  26. [34]

    Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. 2017. Focal loss for dense object detection. In Proceedings of the IEEE international conference on computer vision . 2980–2988

    2017

  27. [35]

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. 2021. Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) . 10012–10022

    2021

  28. [36]

    Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, and Saining Xie. 2022. A ConvNet for the 2020s. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 11976–11986

    2022

  29. [37]

    Tongtong Lu, Julie M Jorns, Dong Hye Ye, Mollie Patton, Renee Fisher, Amanda Emmrich, Taly Gilat Schmidt, Tina Yen, and Bing Yu. 2022. Automated assessment of breast margins in deep ultraviolet fluorescence images using texture analysis. Biomedical Optics Express 13, 9 (2022),...

    2022

  30. [38]

    Ling Ma and Baowei Fei. 2021. Comprehensive review of surgical microscopes: Technology development and medical applications. Journal of Biomedical Optics 26, 1 (2021), 010901–010901

    2021

  31. [39]

    Meena S Moran, Stuart J Schnitt, Armando E Giuliano, Jay R Harris, Seema A Khan, Janet Horton, Suzanne Klimberg, Mariana Chavez-MacGregor, Gary Freedman, Nehmat Houssami, et al. 2014. Society of Surgical Oncology–American Society for Radiation Oncology consensus guideline on m...

    2014

  32. [40]

    Müller and Bernhard Kainz

    Johanna P. Müller and Bernhard Kainz. 2024. Resource-efficient medical image analysis with self-adapting Forward-Forward networks. In Machine Learning in Medical Imaging . Springer Nature Switzerland, 180–190

    2024

  33. [41]

    Ambara R Pradipta, Tomonori Tanei, Koji Morimoto, Kenzo Shimazu, Shinzaburo Noguchi, and Katsunori Tanaka. 2020. Emerging technologies for real-time intraoperative margin assessment in future breast-conserving surgery. Advanced Science 7, 9 (2020), 1901519

    2020

  34. [42]

    Payel Pramanik et al. 2024. DAU-Net: Dual attention-aided U-Net for segmenting tumor in breast ultrasound images. PLoS ONE 19, 5 (2024)

    2024

  35. [43]

    Salma Nurhaliza Pribadi, Igi Ardiyanto, and Nahar Taufiq. 2024. Optimization of segmentation for true lumen, false lumen, and false lumen thrombus in type B aortic dissection using the 2D fine-tuning U-Net method. In Sixth International Conference on Image, Video Processing, a...

    2024

  36. [44]

    Md Mostafijur Rahman, Mustafa Munir, Debesh Jha, Ulas Bagci, and Radu Marculescu. 2024. PP-SAM: Perturbed Prompts for Robust Adaption of Segment Anything Model for Polyp Segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) ....

    2024

  37. [45]

    Emad A Rakha, Cecily Quinn, Yazan A Masannat, Andrew HS Lee, Puay Hoon Tan, Andreas Karakatsanis, Zoltan Tamas Matrai, Salman Husain M Al Shaibani, Salahddin A Gehani, Abeer Shaaban, et al. 2024. Revisiting surgical margins for invasive breast cancer patients treated with brea...

    2024

  38. [46]

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman Rädle, Chloe Rolland, Laura Gustafson, et al. 2024. SAM 2: Segment anything in images and videos. arXiv preprint arXiv:2408.00714 (2024)

    2024 arXiv

  39. [47]

    Abel Reyes-Angulo and Sidike Paheding. 2023. The Forward-Forward Algorithm as a feature extractor for skin lesion classification: A preliminary study. arXiv preprint arXiv:2307.00617 (2023)

    2023 arXiv

  40. [48]

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. 2015. U-Net: Convolutional networks for biomedical image segmentation. In Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Conference, Munich, Germany, October 5–9, 2015, Proceedings,...

    2015

  41. [49]

    Detection of Breast Cancer Lumpectomy Margin with SAM-incorporated Forward-Forward Contrastive Learning 19

    Springer International Publishing, 234–241. Detection of Breast Cancer Lumpectomy Margin with SAM-incorporated Forward-Forward Contrastive Learning 19

  42. [50]

    Chaitanya Ryali et al. 2023. Hiera: A hierarchical vision transformer without the bells-and-whistles. In International Conference on Machine Learning (ICML). PMLR, 29441–29454

    2023

  43. [51]

    Mark T Scimone, Savitri Krishnamurthy, Gopi Maguluri, Dorin Preda, Jesung Park, John Grimble, Min Song, Kechen Ban, and Nicusor Iftimia. 2021. Assessment of breast cancer surgical margins with multimodal optical microscopy: A feasibility clinical study. PLOS One 16, 2 (2021), e0245334

    2021

  44. [52]

    Rebecca L Siegel, Kimberly D Miller, Nikita Sandeep Wagle, and Ahmedin Jemal. 2023. Cancer statistics, 2023. Ca Cancer J Clin 73, 1 (2023), 17–48

    2023

  45. [53]

    Tyrell To, Saba Heidari Gheshlaghi, and Dong Hye Ye. 2022. Deep learning for breast cancer classification of deep ultraviolet fluorescence images toward intra-operative margin assessment. In 2022 44th Annual International Conference of the IEEE Engineering in Medicine & Biolog...

    2022

  46. [54]

    Dinusha Veluponnar, Lisanne L de Boer, Freija Geldof, Lynn-Jade S Jong, Marcos Da Silva Guimaraes, Marie-Jeanne TFD Vrancken Peeters, Frederieke van Duijnhoven, Theo Ruers, and Behdad Dashtbozorg. 2023. Toward intraoperative margin assessment using a deep learning-based approa...

    2023

  47. [55]

    Tyler Ward, Braxton McFarland, Sahar Nozad, Talal Arshad, Hafsa Nebbache, Jin Chen, Xiaoqin Wang, and Abdullah-Al-Zubaer Imran. 2025. Automated Intraoperative Lumpectomy Margin Detection using SAM-Incorporated Forward-Forward Contrastive Learning. In Medical Imaging with Deep ...

    2025

  48. [56]

    Bin Xie, Hao Tang, Bin Duan, Dawen Cai, and Yan Yan. 2024. MaskSAM: Towards Auto-prompt SAM with Mask Classification for Medical Image Segmentation. arXiv preprint arXiv:2403.14103 (2024)

    2024 arXiv

  49. [57]

    Qing Xu, Wenwei Kuang, Zeyu Zhang, Xueyao Bao, Haoran Chen, and Wenting Duan. 2023. SPPNet: A single-point prompt network for nuclei image segmentation. In International Workshop on Machine Learning in Medical Imaging (MLMI) . Springer, 227–236

    2023

  50. [58]

    Yiwen Ye, Yutong Xie, Jianpeng Zhang, Ziyang Chen, Qi Wu, and Yong Xia. 2024. Continual self-supervised learning: Towards universal multi-modal medical data representation learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 1...

    2024

  51. [59]

    Paul A Yushkevich, Yang Gao, and Guido Gerig. 2016. ITK-SNAP: An interactive tool for semi-automatic segmentation of multi-modality biomedical images. In 2016 38th annual international conference of the IEEE engineering in medicine and biology society (EMBC) . IEEE, 3342–3345

    2016

  52. [60]

    Yuxuan Zhang, Tianheng Cheng, Rui Hu, Lei Liu, Heng Liu, Longjin Ran, Xiaoxin Chen, Wenyu Liu, and Xinggang Wang. 2024. EVF-SAM: Early vision-language fusion for text-prompted Segment Anything Model. arXiv preprint arXiv:2406.20076 (2024)

    2024 arXiv

  53. [61]

    Chunpeng Zhou, Kangjie Ning, Qianqian Shen, Sheng Zhou, Zhi Yu, and Haishuai Wang. 2024. SAM-SP: Self-prompting makes sam great again. arXiv preprint arXiv:2408.12364 (2024)

    2024 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.