Pith. sign in

REVIEW 3 major objections 5 minor 31 references

LLM clinical knowledge can classify keloids versus hypertrophic scars by writing local feature programs, not by looking at patient photos.

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 · grok-4.5

2026-07-31 04:16 UTC pith:SEFH4HQT

load-bearing objection Clean methods paper: image-blind LLM feature programs beat foundation probes on a real three-hospital scar task, with honest nesting and real soft spots on site count and search overfitting. the 3 major comments →

arxiv 2607.28538 v1 pith:SEFH4HQT submitted 2026-07-30 cs.CV cs.LG

ScaFE: Data-Efficient Scar Classification with LLM-Generated Clinical Feature Programs

Ruman Wang , Hangting Ye This is my paper
classification cs.CV cs.LG
keywords scar classificationkeloidhypertrophic scarLLM feature engineeringprogram synthesisdata-efficient medical imagingcross-site generalizationauditable clinical AI
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.

Distinguishing keloids from hypertrophic scars in ordinary clinical photos is hard when labeled cases are scarce and hospitals capture images differently. Sending photos to a hosted vision model raises privacy and audit problems, and end-to-end networks need a lot of specialist data. This paper argues that a language model can instead act as a feature engineer: it retrieves clinical sources, writes short deterministic programs that measure visually assessable scar attributes, and revises those programs using only aggregate validation statistics and feature-importance summaries while raw images stay inside the hospital. A simple Random Forest then classifies the resulting table of measurements. On 600 photos from three hospitals, the approach beats strong image baselines under leave-one-site-out testing and keeps a large lead when only a tenth of the development data is available. The point is that clinical knowledge can enter the pipeline as inspectable code rather than as an opaque image diagnosis.

Core claim

Under patient-level leave-one-site-out evaluation on 600 photographs from three hospitals, ScaFE reaches 81.0% site-macro balanced accuracy—10 points above the strongest baseline—by transferring LLM clinical knowledge into deterministic local feature programs rather than direct vision-language diagnosis, and it retains a 11.8-point lead at 10% of the development data.

What carries the argument

ScaFE (Scar Feature Engineering): a validation-guided loop in which a web-enabled LLM generates evidence-linked Python feature programs that run in a restricted local sandbox; only aggregate errors, confusion counts, balanced accuracy, and SHAP summaries return for iterative repair, after which a fixed Random Forest learns the decision boundary on the structured features.

Load-bearing premise

Ordinary clinical photographs, measured by short deterministic programs under a restricted library, carry enough keloid-versus-hypertrophic-scar signal to generalize to an unseen hospital when the language model only sees aggregate validation feedback.

What would settle it

Rerun the same leave-one-site-out protocol on a new hospital or prospective cohort where the true distinction depends on palpation, three-dimensional growth, or history the sandbox cannot measure; if ScaFE then falls to baseline levels while image models hold up, the central claim fails.

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

If this is right

  • Cross-site scar triage can keep photographs inside the hospital while still using hosted LLM knowledge at search time.
  • Low-data medical image tasks that already have named clinical scales can be attacked as feature-program search rather than end-to-end fine-tuning.
  • Every retained feature ships with a source passage and executable definition, so auditors can inspect why a prediction was made without opening the LLM.
  • Refinement with only aggregate SHAP and confusion feedback can raise program executability and evidence coverage without exposing patient-level outputs.

Where Pith is reading between the lines

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

  • The same bounded program-synthesis pattern could transfer to other photo-only dermatology distinctions that clinicians already describe with named visual attributes.
  • If sandbox libraries cannot express the decisive visual cues, the method will plateau even with perfect literature retrieval—suggesting a natural stress test against tasks that need depth or tactile proxies.
  • Governance teams may prefer this split (hosted LLM for code, local RF for decisions) as a template whenever data-residency rules block direct multimodal API calls.

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

3 major / 5 minor

Summary. ScaFE proposes transferring clinical knowledge from a web-enabled LLM into deterministic, locally executed Python feature programs for keloid vs. hypertrophic scar classification, rather than sending images to a VLM or training end-to-end image models. Programs are generated from retrieved clinical evidence, refined over T rounds using only aggregate validation diagnostics (confusion counts, balanced accuracy, invalid/constant rates, global SHAP), then paired with a fixed Random Forest. On 600 unique-patient photographs from three hospitals under nested leave-one-site-out evaluation, ScaFE reports 81.0% site-macro balanced accuracy (+10.0 pp over BiomedCLIP) and retains a large lead at 10% development data (72.0%, +11.8 pp), with refinement raising executability from 66.7% to 95.0% and strong evidence-pass rates after filtering.

Significance. If the gains reflect stable, clinically grounded measurements rather than two-site photographic idiosyncrasies, the work is a meaningful contribution to data-efficient and governance-compatible medical image learning. The separation of roles—LLM as knowledge source, sandboxed code as measurer, lightweight learner as decision boundary—is cleanly formulated (§3–4) and addresses a real tension between hosted multimodal models and hospital data constraints. Strengths include nested patient-level leave-one-site-out, matched partitions, near-chance leakage/permutation controls, ablations that isolate literature grounding and semantically aligned feedback (including shuffled-feedback), and data-efficiency curves that re-run the full search inside each subsample rather than only refitting the classifier. The audit of executability, evidence gating, and SHAP faithfulness is also a useful template for program-synthesis pipelines in medicine.

major comments (3)
  1. [§5.1, Eqs. (3)–(5), Algorithm 1] §5.1 and Eqs. (3)–(5)/Alg. 1: The central cross-site claim rests on three leave-one-site-out folds in which program search repeatedly optimizes on D_val drawn from only the two development hospitals. The manuscript correctly notes that three sites cannot support population-of-hospitals inference, but this understates a load-bearing risk: with M×T=12 candidates and multi-round SHAP/confusion feedback, the loop can amplify any visually measurable cue that separates KD/HS inside the two-site pool (illumination, camera mix, skin-tone/body-site imbalance, labeling style) even when evidence records name clinical constructs. Handcrafted+RF at 61.3% shows the gain is search-driven; ablations remove literature/SHAP but do not break dependence on the two-site val loop. Please add analyses that stress this risk—e.g., reporting retained feature definitions and distributions by site, site-stratified
  2. [§5.6, §4.2–4.3] §5.6 (RQ4) and §4.2–4.3: The claim that LLM knowledge is transferred into clinically meaningful measurements is only weakly evidenced in the main text. Expert assessment of source–feature mappings and visual proxies is deferred to Supplementary A.6, while the main results emphasize BAcc and executability. For the knowledge-transfer story to carry the +10 pp claim, the main paper should characterize the final feature sets (names, operational definitions, top SHAP features, examples of what each measures on images) and show that high-importance dimensions correspond to intended scar attributes rather than acquisition correlates. Without that, “evidence-grounded” largely means bibliographic linkage plus local executability, not demonstrated construct validity.
  3. [§5.2, Table 1] §5.2 baselines: Local VLM-direct uses MedGemma-1.5-4B-IT with label-likelihood normalization and lands at 66.2%, below frozen BiomedCLIP probing. Given that a core motivation is superiority over direct VLM diagnosis (and citation of Shiraishi et al.), the comparison should include a stronger multimodal baseline under the same local/no-train constraint (larger medical VLM if deployable, or multiple prompting/decoding regimes with the same budget as M×T). Otherwise the “rather than direct VLM decisions” contrast risks overstating the margin against a weak instantiation of that paradigm, even though the foundation-model probe comparisons remain informative.
minor comments (5)
  1. [Abstract, §5.6] Abstract vs. body wording on evidence rates is inconsistent across versions (e.g., “verified evidence for 91.7% of the final features” vs. candidate evidence-pass 91.7% with final filtering to 100% coverage). Align terminology so “candidate pass” and “retained coverage” are not conflated.
  2. [Table 1, §5.3] Table 1 in the main text reports only BAcc; macro-F1, AUROC, ECE, and CIs are supplementary. For the primary comparison table, at least CIs on the site-macro average (or the paired improvement CI already mentioned in text) would help readers without opening the supplement.
  3. [Figure 2] Figure 2(b) mixes executability/contract/evidence rates with validation BAcc on one axis; consider dual axes or separate panels so scale differences are clearer.
  4. [§5.1] Implementation cites gpt-4.1-2025-04-14 and access dates in 2026; ensure model IDs, prompts, and archived run artifacts are pinned for reproducibility as claimed in §5.1/A.2.
  5. [§1–2] Minor prose duplicates and spacing artifacts appear (e.g., concatenated words in the introduction and related work). A copy-edit pass would improve readability.

Circularity Check

0 steps flagged

No significant circularity: held-out site performance is not defined by the validation objective, and claims rest on sealed leave-one-site-out comparisons rather than by-construction identities.

full rationale

ScaFE’s load-bearing claim is empirical cross-site balanced accuracy under patient-level leave-one-site-out (Table 1; §5.1–5.3), not a first-principles identity. The derivation chain is: LLM + web evidence → candidate feature programs g → local execution → fixed RF fitted on D_fit → selection by validation BAcc on D_val (Eqs. 3–5, 11, 14; Alg. 1) → freeze g* and h* → single evaluation on sealed D_test. Repeated use of D_val is explicitly labeled an optimization set (“only the sealed test set estimates generalization,” §4.3), which is standard nested model selection, not Eq-X-equals-Eq-Y circularity: the reported 81.0% site-macro BAcc and +10.0 pp vs BiomedCLIP are measured on hospitals never seen during search or fitting. SHAP/confusion feedback guide repair but never expose test labels or patient-level outputs to the LLM (§4.3). Baselines (handcrafted+RF, ResNet, EfficientNet, DINOv3, Derm Foundation, BiomedCLIP, local VLM) and negative controls (random encoder, label permutation) are external comparators on the same partitions. Related-work self-citation (Ye et al. 2024 on prototypes) is not load-bearing for the accuracy claim. No uniqueness theorem, fitted scalar renamed as prediction, or definitional reduction of the test metric to the search objective appears in the equations or evaluation protocol. Overfitting to two-site photographic cues is a generalization risk, not circularity under this pass’s criteria.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 2 invented entities

The central empirical claim rests on standard ML evaluation practice, clinical scar ontology from prior scales, and several method-specific design choices (sandbox interface, aggregate feedback sufficiency, RF as fixed evaluator). No new physical entities; free parameters are search and learner knobs fixed before test, not fitted to the sealed sites.

free parameters (6)
  • M candidates per round = 4
    Number of programs proposed each round; fixed at 4 and multiplies validation budget (MT=12).
  • T refinement rounds = 3
    Search depth for iterative repair; fixed at 3 before test evaluation.
  • LLM sampling temperature = 0.2
    Controls generation stochasticity for gpt-4.1 program synthesis.
  • RF hyperparameters (n_trees, max_depth, min_leaf, feature subsample) = 500 trees, depth 8, min leaf 3, sqrt(Kg) features
    Fixed downstream learner configuration used both as search evaluator and final classifier; not retuned per candidate beyond the shared setting.
  • Input resize resolution for ScaFE = 512×512
    Hand-chosen preprocessing size with aspect-preserving pad; affects all local measurements.
  • fit/val split ratio inside development sites = 80% fit / 20% val
    80/20 patient-wise stratified split supplies all refinement signals; choice affects which programs win.
axioms (5)
  • domain assumption Named visually assessable scar concepts from instruments such as VSS/POSAS can be operationalized as deterministic numeric measurements on 2D RGB photos sufficient for KD vs HS discrimination.
    Stated in Introduction and Problem Formulation; nonvisual scale items are acknowledged as unrecoverable yet the method assumes remaining visual proxies suffice cross-site.
  • ad hoc to paper Aggregate validation diagnostics (confusion counts, BAcc, invalid/constant rates, global SHAP) without patient-level outputs suffice for an LLM to repair and improve feature programs.
    Core of the validation-guided loop in §4.3; ablations support it empirically but it is a design premise of ScaFE.
  • ad hoc to paper A restricted local execution interface (approved libraries, no network/labels/metadata) yields faithful, non-leaky measurements of the intended clinical attributes.
    §4.1–4.2 sandbox contract; required for both governance claim and feature validity.
  • domain assumption Leave-one-site-out on three hospitals with patient-level splits is an adequate test of cross-site generalization for the claimed margins.
    §5.1 explicitly limits inference to sampling within observed sites, not the wider hospital population.
  • standard math Standard supervised learning and bootstrap testing assumptions (i.i.d. patients within site after de-duplication; RF/variance estimates meaningful at N=200/site).
    Used for BAcc, paired bootstrap CIs, and Holm correction in §5.1–5.3.
invented entities (2)
  • ScaFE validation-guided feature-program search loop no independent evidence
    purpose: Organize LLM evidence retrieval, local execution, aggregate-only feedback, evidence gating, and frozen RF into one auditable pipeline.
    The paper’s main methodological object; not a physical entity, but the postulated mechanism that produces the accuracy claims.
  • Evidence-gated executable scar feature programs no independent evidence
    purpose: Serve as the inspectable intermediate representation replacing end-to-end image embeddings or direct VLM labels.
    Generated per fold rather than fixed a priori; utility is internal to the reported experiments unless released and re-used elsewhere.

pith-pipeline@v1.2.0-daily-grok45 · 17778 in / 3860 out tokens · 64518 ms · 2026-07-31T04:16:00.982790+00:00 · methodology

0 comments
read the original abstract

Classifying pathological scars from clinical photographs requires distinguishing keloids from hypertrophic scars despite limited expert-labeled data and substantial acquisition variation across hospitals. End-to-end image models remain data-dependent, whereas sending photographs to a hosted vision-language model (VLM) may conflict with local data-governance requirements and yields decisions that are difficult to reproduce and audit. We introduce ScaFE (Scar Feature Engineering), which transfers clinical knowledge from a large language model (LLM) into deterministic, executable feature programs instead of asking the model to diagnose images. A web-enabled LLM retrieves clinical evidence and synthesizes programs that measure visually assessable scar attributes. Candidate programs execute in a restricted local environment, and only aggregate validation statistics and feature-level SHAP summaries are returned for iterative repair and refinement; raw images and patient-level outputs remain local. A lightweight Random Forest then operates on the resulting structured representation. On 600 photographs from three hospitals under leave-one-site-out evaluation, ScaFE achieves 81.0% site-macro balanced accuracy, exceeding the strongest baseline, BiomedCLIP, by 10.0 percentage points. With only 10% of the development data, ScaFE retains 72.0% balanced accuracy and an 11.8-point lead. Iterative refinement also raises the executable-program rate from 66.7% to 95.0%, with verified evidence for 91.7% of the final features. These results show that LLM knowledge can support data-efficient, cross-site medical image classification through local and auditable feature programs rather than direct VLM decisions.

Figures

Figures reproduced from arXiv: 2607.28538 by Hangting Ye, Ruman Wang.

Figure 1
Figure 1. Figure 1: Overview of ScaFE. An LLM turns source-traceable clinical concepts into deterministic feature programs that process [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Search-loop analysis. (a) Held-out BAcc for ScaFE and its ablations. (b) Executability, contract validity, evidence [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Data efficiency under stratified subsam [PITH_FULL_IMAGE:figures/full_fig_p007_3.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

31 extracted references · 1 canonical work pages

  1. [1]

    Angus and Ferguson, Mark W

    Bayat, Ardeshir and McGrouther, D. Angus and Ferguson, Mark W. J. Skin scarring. BMJ. 2003

    2003

  2. [2]

    Keloids and hypertrophic scars: pathophysiology, classification, and treatment

    Berman, Brian and Maderal, Andrea and Raphael, Briana. Keloids and hypertrophic scars: pathophysiology, classification, and treatment. Dermatologic Surgery. 2017

    2017

  3. [3]

    The Vancouver Scar Scale : an administration tool and its interrater reliability

    Baryza, Mary Jo and Baryza, Gregory A. The Vancouver Scar Scale : an administration tool and its interrater reliability. Journal of Burn Care & Rehabilitation. 1995. doi:10.1097/00004630-199509000-00013

    work page doi:10.1097/00004630-199509000-00013 1995

  4. [4]

    and Tempelman, Fenike R

    Draaijers, Lieneke J. and Tempelman, Fenike R. H. and Botman, Yvonne A. M. and Tuinebreijer, Wim E. and Middelkoop, Esther and Kreis, Robert W. and van Zuijlen, Paul P. M. The Patient and Observer Scar Assessment Scale : a reliable and feasible tool for scar evaluation. Plastic and Reconstructive Surgery. 2004. doi:10.1097/01.PRS.0000122207.28773.56

    arXiv 2004

  5. [5]

    and Ko, Justin and Swetter, Susan M

    Esteva, Andre and Kuprel, Brett and Novoa, Roberto A. and Ko, Justin and Swetter, Susan M. and Blau, Helen M. and Thrun, Sebastian. Dermatologist-level classification of skin cancer with deep neural networks. Nature. 2017. doi:10.1038/nature21056

    doi:10.1038/nature21056 2017

  6. [6]

    Deep residual learning for image recognition

    He, Kaiming and Zhang, Xiangyu and Ren, Shaoqing and Sun, Jian. Deep residual learning for image recognition. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2016

    2016

  7. [7]

    EfficientNet : Rethinking model scaling for convolutional neural networks

    Tan, Mingxing and Le, Quoc V. EfficientNet : Rethinking model scaling for convolutional neural networks. International Conference on Machine Learning. 2019

    2019

  8. [8]

    a rli, Nathanael and Chowdhery, Aakanksha and Mansfield, Philip and Demner-Fushman, Dina and Ag \

    Singhal, Karan and Azizi, Shekoofeh and Tu, Tao and Mahdavi, S. Sara and Wei, Jason and Chung, Hyung Won and Scales, Nathan and Tanwani, Ajay and Cole-Lewis, Heather and Pfohl, Stephen and Payne, Perry and Seneviratne, Martin and Gamble, Paul and Kelly, Chris and Babiker, Abubakr and Sch \"a rli, Nathanael and Chowdhery, Aakanksha and Mansfield, Philip an...

    2023

  9. [9]

    Capabilities of GPT-4 on medical challenge problems

    Nori, Harsha and King, Nicholas and McKinney, Scott Mayer and Carignan, Dean and Horvitz, Eric. Capabilities of GPT-4 on medical challenge problems. arXiv preprint arXiv:2303.13375. 2023

    Pith/arXiv arXiv 2023

  10. [10]

    Liu, Yuan and Jain, Ayush and Eng, Clara and Way, David H. and Lee, Kang and Bui, Peggy and Kanada, Kimberly and de Oliveira Marinho, Guilherme and Gallegos, Jessica and Gabriele, Sara and Gupta, Vishakha and Singh, Nalini and Natarajan, Vivek and Hofmann-Wellenhof, Rainer and Corrado, Greg S. and Peng, Lily H. and Webster, Dale R. and Ai, Dennis and Huan...

    2020

  11. [11]

    Pre-trained multimodal large language model enhances dermatological diagnosis using

    Zhou, Juexiao and He, Xiaonan and Sun, Liyuan and Xu, Jiannan and Chen, Xiuying and Chu, Yuetan and Zhou, Longxi and Liao, Xingyu and Zhang, Bin and Afvari, Shawn and Gao, Xin , journal=. Pre-trained multimodal large language model enhances dermatological diagnosis using. 2024 , doi=

    2024

  12. [12]

    Representation learning: A review and new perspectives

    Bengio, Yoshua and Courville, Aaron and Vincent, Pascal. Representation learning: A review and new perspectives. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2013

    2013

  13. [13]

    A few useful things to know about machine learning

    Domingos, Pedro. A few useful things to know about machine learning. Communications of the ACM. 2012

    2012

  14. [14]

    Neurosymbolic AI : The 3rd wave

    Garcez, Artur d'Avila and Lamb, Luis C. Neurosymbolic AI : The 3rd wave. Artificial Intelligence Review. 2023. doi:10.1007/s10462-023-10448-w

    doi:10.1007/s10462-023-10448-w 2023

  15. [15]

    Concept bottleneck models

    Koh, Pang Wei and Nguyen, Thao and Tang, Yew Siang and Mussmann, Stephen and Pierson, Emma and Kim, Been and Liang, Percy. Concept bottleneck models. International Conference on Machine Learning. 2020

    2020

  16. [16]

    Knowledge-based collaborative deep learning for benign-malignant lung nodule classification on chest CT

    Xie, Yutong and Xia, Yong and Zhang, Jianpeng and Song, Yang and Feng, Dagan and Fulham, Michael and Cai, Weidong. Knowledge-based collaborative deep learning for benign-malignant lung nodule classification on chest CT. IEEE Transactions on Medical Imaging. 2019

    2019

  17. [17]

    2023 , doi=

    Li, Chunyuan and Wong, Cliff and Zhang, Sheng and Usuyama, Naoto and Liu, Haotian and Yang, Jianwei and Naumann, Tristan and Poon, Hoifung and Gao, Jianfeng , journal=. 2023 , doi=

    2023

  18. [18]

    Aesthetic Plastic Surgery , volume=

    The potential of chat-based artificial intelligence models in differentiating between keloid and hypertrophic scars: a pilot study , author=. Aesthetic Plastic Surgery , volume=. 2024 , doi=

    2024

  19. [19]

    Random forests

    Breiman, Leo. Random forests. Machine learning. 2001

    2001

  20. [20]

    Medical image analysis , volume=

    A survey on deep learning in medical image analysis , author=. Medical image analysis , volume=. 2017 , doi=

    2017

  21. [21]

    2024 , url=

    Ye, Hangting and Fan, Wei and Song, Xiaozhuang and Zheng, Shun and Zhao, He and Guo, Dandan and Chang, Yi , booktitle=. 2024 , url=

    2024

  22. [22]

    A Unified Approach to Interpreting Model Predictions , url =

    Lundberg, Scott M and Lee, Su-In , booktitle =. A Unified Approach to Interpreting Model Predictions , url =

  23. [23]

    and Naumann, Tristan and Wang, Sheng and Poon, Hoifung , title =

    Zhang, Sheng and Xu, Yanbo and Usuyama, Naoto and Xu, Hanwen and Bagga, Jaspreet and Tinn, Robert and Preston, Sam and Rao, Rajesh and Wei, Mu and Valluri, Naveen and Wong, Cliff and Tupini, Andrea and Wang, Yu and Mazzola, Matt and Shukla, Swadheen and Liden, Lars and Gao, Jianfeng and Crabtree, Angela and Piening, Brian and Bifulco, Carlo and Lungren, M...

    2025

  24. [24]

    Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages =

    Chen, Tianqi and Guestrin, Carlos , title =. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages =. 2016 , doi =

    2016

  25. [25]

    arXiv preprint arXiv:2508.10104 , year =

    Sim. arXiv preprint arXiv:2508.10104 , year =

    Pith/arXiv arXiv

  26. [26]

    Kiraly, Atilla P. and Baur, Sebastien and Philbrick, Kenneth and Mahvar, Fereshteh and Yatziv, Liron and Chen, Tiffany and Sterling, Bram and George, Nick and Jamil, Fayaz and Tang, Jing and Bailey, Kai and Ahmed, Faruk and Goel, Akshay and Ward, Abbi and Yang, Lin and Sellergren, Andrew and Matias, Yossi and Hassidim, Avinatan and Shetty, Shravya and Gol...

    Pith/arXiv arXiv

  27. [27]

    and Kirmizibayrak, Can and Golden, Daniel , title =

    Sellergren, Andrew and Gao, Chufan and Mahvar, Fereshteh and Kohlberger, Timo and Jamil, Fayaz and Traverse, Madeleine and Tono, Alberto and Sadjad, Bashir and Yang, Lin and Lau, Charles and Yatziv, Liron and Chen, Tiffany and Sterling, Bram and Philbrick, Kenneth and Tiwari, Richa and Liu, Yun and Jajoo, Madhuram and Sankarapu, Chandrashekar and Vispute,...

    Pith/arXiv arXiv

  28. [28]

    Peter and Ge, Zongyuan , title =

    Yan, Siyuan and Yu, Zhen and Primiero, Clare and Vico-Alonso, Cristina and Wang, Zhonghua and Yang, Litao and Tschandl, Philipp and Hu, Ming and Ju, Lie and Tan, Gin and Tang, Vincent and Ng, Aik Beng and Powell, David and Bonnington, Paul and See, Simon and Magnaterra, Elisabetta and Ferguson, Peter and Nguyen, Jennifer and Guitera, Pascale and Banuls, J...

    2025

  29. [29]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages =

    Yan, Siyuan and Hu, Ming and Jiang, Yiwen and Li, Xieji and Fei, Hao and Tschandl, Philipp and Kittler, Harald and Ge, Zongyuan , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages =

  30. [30]

    and Hla, Michael and Moor, Michael and Adithan, Subathra and Kwak, Stephen and Topol, Eric J

    Rao, Vishwanatha M. and Hla, Michael and Moor, Michael and Adithan, Subathra and Kwak, Stephen and Topol, Eric J. and Rajpurkar, Pranav , title =. Nature , volume =. 2025 , doi =

    2025

  31. [31]

    and Diao, James A

    Buckley, Thomas A. and Diao, James A. and Srivastava, Cam N. and Brodeur, Peter G. and Rajpurkar, Pranav and Rodman, Adam and Manrai, Arjun K. , title =. Nature Communications , volume =. 2026 , doi =

    2026