Pith. sign in

REVIEW 5 major objections 5 minor 35 references

CheXLearner: Text-Guided Fine-Grained Representation Learning for Progression Detection

T0 review · 5 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read CheXLearner's text-guided hyperbolic alignment detects region-level chest X-ray progression at 81.12% accuracy.

desk verdict A believable method with a real evaluation gap: the +17-point gain over CheXDetector is internally consistent, but the label-derived text supervision and the unverified patient-level split both need referee attention. read the letter →

arxiv 2505.06903 v1 pith:SHJOUOU7 submitted 2025-05-11 cs.CV

Yuanzhuo Wang , Junwen Duan , Xinyu Li , Jianxin Wang This is my paper · ORCID
classification cs.CV
keywords diseaseprogressiondetectionchestX-rayvision-languagerepresentationlearninghyperbolicmanifoldalignmentDETRtemporalmedicalimagingChest-ImaGenome
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

CheXLearner tries to establish that chest X-ray disease progression can be detected far more accurately when the model is supervised, at the level of each anatomical region, by text describing what changed between two images. It builds an end-to-end pipeline that detects 12 anatomical regions with DETR, aligns the two time points' structures on a hyperbolic manifold to remove imaging-angle and breathing differences, and uses region-level progression descriptions to guide the visual features. On Chest-ImaGenome the paper reports 81.12% average accuracy and 80.32% F1 on the improved/worsened/no-change task across 12 regions, beating the previous region-level system by 17.2 accuracy points and 11.05 F1 points. The larger message is that fine-grained text guidance plus geometric alignment, trained end-to-end, is a viable recipe for temporal medical image analysis.

What carries the argument

The carrying object is Med-MAM, a medical manifold alignment module. It takes the same anatomical region's features from the two time points, builds an explicit difference vector and a contrastive context in Euclidean space, maps both images' features into a Poincaré disk with learnable curvature, computes the Riemannian log map between them, and parallel-transports the resulting discrepancy vector to the target image's tangent space. This transport is what removes viewpoint and breathing differences while preserving pathological change. The aligned features are compressed to a shared dimension, then pulled toward MedCLIP-encoded progression descriptions by an InfoNCE contrastive loss and an image-text matching loss, with class-weighted cross-entropy for the three-way progression decision; the whole pipeline trains together with DETR unfrozen.

What would settle it

Re-run the progression benchmark with the region-to-text description pairs randomly shuffled while keeping the ground-truth labels fixed; if accuracy stays near 81%, the paper's claim that text semantics drives the gain is falsified, because only the region token or the template would remain informative.

Watch

Extended reading notes

Core claim

The central claim is that regional semantic text is the missing supervision for temporal chest X-ray analysis: instead of aligning whole images to whole reports, CheXLearner detects each anatomical region, aligns the two images' versions of that region on a Riemannian hyperbolic manifold, and pulls the resulting visual representation toward the region's textual progression description. With this design the paper reports 81.12% average accuracy and 80.32% F1 on progression detection across 12 anatomical regions, gains of +17.2 and +11.05 over CheXDetector, and a 91.52% average AUC on downstream disease classification. The ablation attributes the largest share of the gain to unfreezing DETR (F1 +7.22), then to the contrastive text loss, with the manifold fusion adding roughly another 2 points of accuracy.

Load-bearing premise

The result rests on the automatically generated regional progression labels and text descriptions being accurate, and on the data split keeping each patient's pairs in one partition; if either fails, the reported +17.2-point gain could be inflated.

Editorial extensions

If this is right

  • Region-level text descriptions become a scalable supervision source: any dataset with radiology reports can be mined into per-region progression descriptions, removing the need for manually drawn progression labels.
  • End-to-end training with the detector unfrozen should be the default design in longitudinal medical imaging, since the ablation attributes the largest single F1 gain (+7.22) to unfreezing DETR.
  • Hyperbolic alignment with parallel transport generalizes to any paired anatomical data, so CT follow-up, MRI longitudinal, and even histology time-series could inherit the same discrepancy-feature construction.
  • The downstream disease-classification AUC gain (+1.86 points on average) implies the same regional text-supervised features are transferable rather than overfit to the progression task.

Reading between the lines

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

  • A natural audit not reported in the paper: because 318,373 of 427,752 region pairs lack real descriptions and receive a fixed 'no evident change' template, the template may encode the majority class and inflate the contrastive gain; shuffling text assignments would reveal this.
  • The Chest-ImaGenome silver labels come from automatically extracted rules rather than expert review; if they contain systematic errors in the same regions where DETR detection is weakest (costophrenic angles at 73–81% average precision), part of the apparent improvement there could be learning to mimic label noise.
  • If the gains replicate, a cheap extension is to train with real descriptions only and serve template text for unseen patients, or to distill the text-aligned visual model into a text-free student for deployment where reports are unavailable.
  • The parallel-transport discrepancy vector is essentially a geometric 'changed since last visit' feature; it could be combined with time-interval embeddings to predict not just direction (better/worse/stable) but magnitude and rate of change.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes CheXLearner, an end-to-end framework for detecting disease progression in temporal chest X-ray image pairs at the level of 12 anatomical regions. The method uses a pre-trained DETR detector (fine-tuned end-to-end) to extract region features, a hyperbolic manifold module (Med-MAM) with parallel transport to align temporal structure differences, and regional progression text descriptions as auxiliary supervision via image-text contrastive (ITC) and image-text matching (ITM) tasks. The paper reports 81.12% average accuracy and 80.32% F1 on progression detection, outperforming CheXDetector by +17.2%/+11.05%, and 91.52% average AUC on downstream disease classification. Ablation studies attribute the gains to unfreezing DETR and to the text contrastive loss.

Significance. If the empirical claims are valid, CheXLearner is a potentially useful contribution to region-level temporal chest X-ray analysis: it is one of the few methods that combine detection, geometric alignment, and text-guided representation learning in an end-to-end pipeline. The ablations are informative: Table 6 shows that unfreezing DETR improves F1 by 7.22 percentage points, and ITC adds a further large gain; Table 7 shows that the proposed fusion module outperforms simpler alternatives. The method is based on public Chest-ImaGenome data and is reproducible in principle, though no code is provided. However, the significance is conditional on resolving evaluation-robustness issues: the absence of a patient-stratified split, the aggressive labeling of 318,373 unannotated region pairs as 'no change', and the lack of error bars mean the headline numbers may not be reliable as stated.

major comments (5)
  1. [§4.1] The paper does not state that the 70/10/20 split is patient-stratified. Chest-ImaGenome is derived from MIMIC-CXR, which contains multiple images per patient, and the unit of analysis is a temporal CXR pair. Without a patient-level split, the same patient's images can appear in both training and test sets, allowing the model to memorize patient-specific appearance (e.g., body habitus, device artifacts) instead of learning general progression patterns. This could directly inflate the reported +17.2% accuracy gain over CheXDetector. The authors should specify whether the split is by patient ID and, if it is not, rerun the evaluation with a patient-stratified split and report the resulting numbers.
  2. [§4.1] All 318,373 region pairs without extracted progression descriptions are assigned the healthy 'no change' label and a template text. These samples account for roughly 75% of the 423,672 region pairs in the progression dataset. If the absence of a description reflects annotation incompleteness rather than a true absence of change, then the ground-truth labels themselves are contaminated. Because a large fraction of the test set is affected, the reported accuracy and F1 values are not interpretable without further analysis. Please report per-class precision/recall and confusion matrices, and add a sensitivity analysis in which the unannotated samples are excluded or labeled using an independent criterion.
  3. [§3.6] The total loss in Eq. (14) combines L_detr, L_contrast, and L_cls, but the method description and ablation study include an Image-Text Matching (ITM) task (Figure 2 and Table 6). No ITM loss is ever defined. This omission makes the method non-reproducible and leaves the ITM ablation rows without a clear interpretation. Please define the ITM objective and state how it enters Eq. (14).
  4. [§3.4/§4.1] The text supervision is not independent of the progression labels. The regional progression descriptions are extracted from the same Chest-ImaGenome scene-graph annotation pipeline that produces the progression labels, and the 'no change' template is assigned using the same absence criterion. If the text merely restates the progression outcome or a generic healthy template, then L_contrast is a label-derived auxiliary loss rather than a genuine cross-modal semantic signal. Please show representative examples of the actual text descriptions used, and test whether the text contributes beyond the class labels, for example by comparing against a condition that replaces text with a one-hot class encoding.
  5. [§4.4] All results come from a single run without error bars or significance tests. The headline differences (Table 3, +17.2% accuracy) are large, but the ablation differences are small (Table 7, 2.03% between Med-MAM and Transformer) and could be within run-to-run noise. Please report means and standard deviations over at least three independent runs, or provide a valid statistical test for the key comparisons.
minor comments (5)
  1. [Table 1] The row 'cardiac silhouette' appears twice with identical counts, making the table list 13 rows instead of 12 anatomical locations; the totals correspond to the 12 unique rows. Please remove the duplicate.
  2. [§3.3.1] After Eq. (3), the fused feature fe is described as fe∈R, but since fe = α·c + (1−α)·Δe with c, Δe ∈ R^{3d}, the correct statement is fe ∈ R^{3d}.
  3. [§3.3.2] Equation (8), the parallel transport formula, should be referenced to a standard source; as written, it does not match the standard Poincaré parallel transport formula found in the literature. Please verify the expression and add a citation.
  4. [§4.1] The dataset description says 9 target diseases and 12 anatomical regions, but the downstream evaluation in Table 5 reports AUC only for CheXDetector and CheXLearner; the downstream performance of the other baselines is not given. Please clarify whether downstream classification was evaluated only on these two models.
  5. [Throughout] There are occasional typos and style inconsistencies, e.g., 'ChexDetector' versus 'CheXDetector' and the phrase 'in an end-to-end training framework' is repeated. Please proofread.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper reports an empirical end-to-end training method benchmarked against external baselines, and no prediction reduces to its inputs by construction.

full rationale

CheXLearner is an empirical end-to-end training method rather than a formal derivation. The reported progression detection accuracy is obtained by optimizing the total loss L_total = L_detr + L_contrast + L_cls (Eq. 14) and evaluating on a held-out test split; no equation re-inserts a fitted parameter or training text as the test-time prediction. The regional text descriptions enter only through the contrastive (ITC) and matching (ITM) training losses (Eqs. 11-12), and the test-time model uses visual features only, so the predictions are not forced by construction to equal the text or the labels. The fact that both the progression labels and the text descriptions originate from the Chest-ImaGenome annotation pipeline is a data-dependence caveat about possible supervision leakage, not a circular derivation. The assignment of 318,373 no-description samples to a healthy 'no change' template is a labeling heuristic that could affect label quality, but it does not make the reported benchmark equivalent to the training inputs by construction. Comparisons to CheXDetector, CheXRelNet, and CheXRelFormer are external benchmarks, and no load-bearing self-citation chain is present. The skeptic's patient-split concern is a correctness and data-leakage risk, not a circularity argument. The acknowledged limitation in Section 5 that descriptions are 'conclusion-oriented with insufficient detail' is a stated limitation, not a circular step. Therefore, the paper shows no significant circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The method introduces a learned module, Med-MAM, rather than a new physical or conceptual entity, so no invented entities are listed. The main external inputs are the pre-trained DETR and MedCLIP text encoder, the Chest-ImaGenome labels and scene graphs, and the chosen manifold and contrastive hyperparameters.

free parameters (3)
  • Poincaré curvature c = learnable, initialized to 0.1
    The manifold curvature is trained with the network and is central to Med-MAM; no sensitivity analysis is reported (Section 3.3.2).
  • InfoNCE temperature tau = 0.05
    Set by hand for the contrastive loss; a standard value, but no tuning analysis is given (Section 3.6).
  • Loss weighting lambda = 1.0 for all loss terms
    All losses are equally weighted without ablation, so the balance between detection, contrastive, and classification losses is a chosen parameter (Section 3.6).
assumptions (4)
  • domain assumption DETR bounding boxes and decoder features from layers 1, 3, and 5 reliably identify the same anatomical region across temporal image pairs.
    The entire pipeline is built on DETR outputs; the costophrenic angle reaches only 73% to 81% AP, so region feature extraction is fragile there (Section 4.4.1).
  • domain assumption Text descriptions extracted from Chest-ImaGenome scene graphs, plus the healthy template, provide correct regional progression semantics for contrastive learning.
    No manual verification is reported, and 318,373 region pairs without descriptions are labeled as healthy 'no change', which can bias labels and supervision (Section 4.1).
  • standard math The Poincaré ball with learnable curvature and the parallel transport formula in Eq. (8) correctly align temporal anatomical structures and preserve pathological discrepancy vectors.
    The paper invokes standard Riemannian geometry but gives no derivation or numerical validation of Eq. (8), and the projection in Eq. (5) is not the typical Poincaré projection.
  • domain assumption Chest-ImaGenome silver-standard labels are reliable enough to support a claimed +17.2% accuracy improvement.
    The silver-standard labels are generated by rule-based NLP and atlas-based detection rather than radiologist review; label noise could inflate or suppress apparent model differences (Section 4.1).

how reviews work

0 comments
Cite this review

Pith. "Pith review of CheXLearner: Text-Guided Fine-Grained Representation Learning for Progression Detection." pith.science (2026). https://pith.science/paper/SHJOUOU7

@misc{pith2026250506903,
  author       = {Pith},
  title        = {Pith review of: CheXLearner: Text-Guided Fine-Grained Representation Learning for Progression Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SHJOUOU7}},
  note         = {Machine review of arXiv:2505.06903}
}
read the original abstract

Temporal medical image analysis is essential for clinical decision-making, yet existing methods either align images and text at a coarse level - causing potential semantic mismatches - or depend solely on visual information, lacking medical semantic integration. We present CheXLearner, the first end-to-end framework that unifies anatomical region detection, Riemannian manifold-based structure alignment, and fine-grained regional semantic guidance. Our proposed Med-Manifold Alignment Module (Med-MAM) leverages hyperbolic geometry to robustly align anatomical structures and capture pathologically meaningful discrepancies across temporal chest X-rays. By introducing regional progression descriptions as supervision, CheXLearner achieves enhanced cross-modal representation learning and supports dynamic low-level feature optimization. Experiments show that CheXLearner achieves 81.12% (+17.2%) average accuracy and 80.32% (+11.05%) F1-score on anatomical region progression detection - substantially outperforming state-of-the-art baselines, especially in structurally complex regions. Additionally, our model attains a 91.52% average AUC score in downstream disease classification, validating its superior feature representation.

Figures

Figures reproduced from arXiv: 2505.06903 by the authors.

Figure 1
Figure 1. (a) Methods (e.g., MLRG [17], MLIP [15]) that rely on coarse image-report alignment. (b) Methods (e.g., CheXDetector) [5] that detect anatomical region progression but ignore semantic integration and suffer from regional noise. (c) Our proposed end-to-end model with anatomical structure alignment and regional progression semantics integration. imaging perspectives and physiological motion, particularly in fine￾grain… view at source ↗
Figure 2
Figure 2. Overview of the CheXLearner framework. The model extracts region-of-interest (ROI) features from temporal CXRs using a DETR pretrained on CXR data. To address anatomical structure misalignment, the Med-MAM module mitigates domain shifts on Manifold and generates discrepancy features. Feature learning is facilitated through fine-grained image-text contrastive (ITC) and image-text matching (ITM) tasks that jointly mod… view at source ↗
Figure 3
Figure 3. Euclidean Dynamic Context-Difference Fusion module. As illustrated in 3, this component constructs dynamic fusion of base features and difference features within Euclidean space. The input features f (1,3,5) 1 ,f (1,3,5) 2 ∈ R 3d correspond to feature repre￾sentations of the same anatomical region extracted from DETR’s 1,3,5 decoder layers for image pairs (X, X′ )i. The process begins with difference enhancement: th… view at source ↗
Figures from the paper (1 more)
Figure 5
Figure 5. Figure 5: Scatter Plot of Feature Distributions for Our CheXLearner (upper region) vs. CheXDetector (lower region) [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 29 canonical work pages

  1. [1]

    Amodio and S

    M. Amodio and S. Krishnaswamy. MAGAN: Aligning biological man- ifolds. In J. Dy and A. Krause, editors,Proceedings of the 35th Interna- tional Conference on Machine Learning, volume 80 of Proceedings of Machine Learning Research, pages 215–223. PMLR, 10–15 Jul 2018. URL https://proceedings.mlr.press/v80/amodio18a.html

    work page 2018

  2. [2]

    Bannur, S

    S. Bannur, S. Hyland, Q. Liu, F. Perez-Garcia, M. Ilse, D. C. Castro, B. Boecking, H. Sharma, K. Bouzid, A. Thieme, et al. Learning to exploit temporal structure for biomedical vision-language processing. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15016–15027, 2023

    work page 2023

  3. [3]

    Carion, F

    N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko. End-to-end object detection with transformers. In Euro- pean conference on computer vision, pages 213–229. Springer, 2020

    work page 2020

  4. [4]

    Q. Chen, Q. Fu, H. Bai, and Y . Hong. Longformer: longitudinal trans- former for alzheimer’s disease classification with structural mris. In Proceedings of the IEEE/CVF winter conference on applications of computer vision, pages 3575–3584, 2024

    work page 2024

  5. [5]

    Eshraghi Dehaghani, A

    M. Eshraghi Dehaghani, A. Sabour, A. B. Madu, I. Lourentzou, and M. Moradi. Representation learning with a transformer-based detection model for localized chest x-ray disease and progression detection. In International Conference on Medical Image Computing and Computer- Assisted Intervention, pages 578–587. Springer, 2024

    work page 2024

  6. [6]

    Everingham, S

    M. Everingham, S. A. Eslami, L. Van Gool, C. K. Williams, J. Winn, and A. Zisserman. The pascal visual object classes challenge: A retro- spective. International journal of computer vision, 111:98–136, 2015

    work page 2015

  7. [7]

    Z. Guo, J. Ma, Y . Xu, Y . Wang, L. Wang, and H. Chen. Histgen: Histopathology report generation via local-global feature encoding and cross-modal context interaction. In International Conference on Medi- cal Image Computing and Computer-Assisted Intervention, pages 189–

    work page

  8. [8]

    Hanik, H.-C

    M. Hanik, H.-C. Hege, and C. v. Tycowicz. A nonlinear hierarchical model for longitudinal data on manifolds. In 2022 IEEE 19th Interna- tional Symposium on Biomedical Imaging (ISBI), pages 1–5, 2022. doi: 10.1109/ISBI52829.2022.9761465

    arXiv 2022

Show all 35 references
  1. [9]

    Huang, Y

    H. Huang, Y . Huang, S. Xie, L. Lin, R. Tong, Y .-W. Chen, Y . Li, and Y . Zheng. Combinatorial cnn-transformer learning with manifold con- straints for semi-supervised medical image segmentation. Proceedings of the AAAI Conference on Artificial Intelligence , 38(3):2330–2338,...

    2024 doi

  2. [10]

    A. E. Johnson, T. J. Pollard, S. J. Berkowitz, N. R. Greenbaum, M. P. Lungren, C.-y. Deng, R. G. Mark, and S. Horng. Mimic-cxr, a de- identified publicly available database of chest radiographs with free-text reports. Scientific data, 6(1):317, 2019

    2019

  3. [11]

    Karwande, A

    G. Karwande, A. B. Mbakwe, J. T. Wu, L. A. Celi, M. Moradi, and I. Lourentzou. Chexrelnet: An anatomy-aware model for tracking lon- gitudinal relationships between chest x-rays. In International Confer- ence on Medical Image Computing and Computer-Assisted Interven- tion, page...

    2022

  4. [12]

    N. Konz, H. Gu, H. Dong, and M. A. Mazurowski. The intrinsic man- ifolds of radiological images and their role in deep learning. In In- ternational Conference on Medical Image Computing and Computer- Assisted Intervention, pages 684–694. Springer, 2022

    2022

  5. [13]

    J. Lee, W. Yoon, S. Kim, D. Kim, S. Kim, C. H. So, and J. Kang. Biobert: a pre-trained biomedical language representation model for biomedical text mining. Bioinformatics, 36(4):1234–1240, 2020

    2020

  6. [14]

    Y . Li, Z. Wang, Y . Liu, L. Wang, L. Liu, and L. Zhou. Kar- gen: Knowledge-enhanced automated radiology report generation us- ing large language models. In International Conference on Medical Image Computing and Computer-Assisted Intervention, pages 382–392. Springer, 2024

    2024

  7. [15]

    Z. Li, L. T. Yang, B. Ren, X. Nie, Z. Gao, C. Tan, and S. Z. Li. Mlip: Enhancing medical visual representation with divergence encoder and knowledge-guided contrastive learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 11704–...

    2024

  8. [16]

    K. Liu, Z. Ma, X. Kang, Z. Zhong, Z. Jiao, G. Baird, H. Bai, and Q. Miao. Structural entities extraction and patient indications incorpo- ration for chest x-ray report generation. In International Conference on Medical Image Computing and Computer-Assisted Intervention , pages...

    2024

  9. [17]

    K. Liu, Z. Ma, X. Kang, Y . Li, K. Xie, Z. Jiao, and Q. Miao. Enhanced contrastive learning with multi-view longitudinal data for chest x-ray report generation. arXiv preprint arXiv:2502.20056, 2025

    2025

  10. [18]

    Y . Luo, H. Li, X. Wu, M. Cao, X. Huang, Z. Zhu, P. Liao, H. Chen, and Y . Zhang. Textual inversion and self-supervised refinement for ra- diology report generation. In International Conference on Medical Im- age Computing and Computer-Assisted Intervention , pages 681–691. Sp...

    2024

  11. [19]

    A. B. Mbakwe, L. Wang, M. Moradi, and I. Lourentzou. Hierarchical vision transformers for disease progression detection in chest x-ray im- ages. In International Conference on Medical Image Computing and Computer-Assisted Intervention, pages 685–695. Springer, 2023

    2023

  12. [20]

    D. Y . Oh, J. Kim, and K. J. Lee. Longitudinal change detection on chest x-rays using geometric correlation maps. In Medical Image Computing and Computer Assisted Intervention–MICCAI 2019: 22nd International Conference, Shenzhen, China, October 13–17, 2019, Proceedings, Part V...

    2019

  13. [21]

    A. v. d. Oord, Y . Li, and O. Vinyals. Representation learning with con- trastive predictive coding. arXiv preprint arXiv:1807.03748, 2018

    2018 arXiv

  14. [22]

    X. Pennec. Intrinsic statistics on riemannian manifolds: Basic tools for geometric measurements. Journal of Mathematical Imaging and Vision, 25:127–154, 2006

    2006

  15. [23]

    Santeramo, S

    R. Santeramo, S. Withey, and G. Montana. Longitudinal detection of radiological abnormalities with time-modulated lstm. In Deep learning in medical image analysis and multimodal learning for clinical deci- sion support: 4th international workshop, DLMIA 2018, and 8th inter- na...

    2018

  16. [24]

    S. Shen, J. Cao, Y . Yin, and R. Zimmermann. Manifold-aware local fea- ture modeling for semi-supervised medical image segmentation. arXiv preprint arXiv:2410.10287, 2024

    2024 arXiv

  17. [25]

    Singh, M

    R. Singh, M. K. Kalra, C. Nitiwarangkul, J. A. Patti, F. Homayounieh, A. Padole, P. Rao, P. Putha, V . V . Muse, A. Sharma, et al. Deep learning in chest radiography: detection of findings and presence of change.PloS one, 13(10):e0204155, 2018

    2018

  18. [26]

    Sriram, M

    A. Sriram, M. Muckley, K. Sinha, F. Shamout, J. Pineau, K. J. Geras, L. Azour, Y . Aphinyanaphongs, N. Yakubova, and W. Moore. Covid-19 prognosis via self-supervised representation learning and multi-image prediction. arXiv preprint arXiv:2101.04909, 2021

    2021 arXiv

  19. [27]

    Y . Sun, K. Li, D. Chen, Y . Hu, and S. Zhang. Lomia-t: A transformer- based longitudinal medical image analysis framework for predicting treatment response of esophageal cancer. In International Confer- ence on Medical Image Computing and Computer-Assisted Interven- tion, pag...

    2024

  20. [28]

    Wang and S

    C. Wang and S. Mahadevan. Manifold alignment preserving global ge- ometry. In IJCAI, pages 1743–1749, 2013

    2013

  21. [29]

    Z. Wang, Z. Wu, D. Agarwal, and J. Sun. Medclip: Contrastive learning from unpaired medical images and text. In Proceedings of the Confer- ence on Empirical Methods in Natural Language Processing. Confer- ence on Empirical Methods in Natural Language Processing , volume 2022, ...

    2022

  22. [30]

    J. T. Wu, N. N. Agu, I. Lourentzou, A. Sharma, J. A. Paguio, J. S. Yao, E. C. Dee, W. Mitchell, S. Kashyap, A. Giovannini, et al. Chest imagenome dataset for clinical reasoning. arXiv preprint arXiv:2108.00316, 2021

    2021 arXiv

  23. [31]

    Y . Xu, A. Hosny, R. Zeleznik, C. Parmar, T. Coroller, I. Franco, R. H. Mak, and H. J. Aerts. Deep learning predicts lung cancer treatment response from serial medical imaging. Clinical Cancer Research , 25 (11):3266–3275, 2019

    2019

  24. [32]

    W. Yu, N. Xu, N. Huang, and H. Chen. Bridging the gap: Geometry- centric discriminative manifold distribution alignment for enhanced classification in colorectal cancer imaging. Computers in Biology and Medicine, 170:107998, 2024

    2024

  25. [33]

    Zeghlache, P.-H

    R. Zeghlache, P.-H. Conze, M. El Habib Daho, Y . Li, H. Le Boité, R. Ta- dayoni, P. Massin, B. Cochener, A. Rezaei, I. Brahim, et al. Latim: Longitudinal representation learning in continuous-time models to pre- dict disease progression. In International Conference on Medical ...

    2024

  26. [34]

    X. Zeng, B. Lu, and J. Zhang. Medical image synthesis algorithm based on vision graph neural network with manifold matching. Biomedical Signal Processing and Control, 103:107381, 2025

    2025

  27. [35]

    Z. Zhu, X. Cheng, Y . Zhang, Z. Chen, Q. Long, H. Li, Z. Huang, X. Wu, and Y . Zheng. Multivariate cooperative game for image-report pairs: Hierarchical semantic alignment for medical report generation. In In- ternational Conference on Medical Image Computing and Computer- Ass...

    2024

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.