pith. sign in

arxiv: 2604.23335 · v1 · submitted 2026-04-25 · 💻 cs.CV

H-SemiS: Hierarchical Fusion of Semi and Self-Supervised Learning for Knee Osteoarthritis Severity Grading

Chandravardhan Singh Raghaw , Anushka Parwal , Shahid Shafi Dar , Prajakta Darade , Nagendra Kumar This is my paper

Pith reviewed 2026-05-08 08:42 UTC · model grok-4.3

classification 💻 cs.CV
keywords knee osteoarthritisseverity gradingsemi-supervised learningself-supervised learningteacher-student modelmedical image analysisclass imbalanceX-ray radiographs
0
0 comments X

The pith

A hierarchical decomposition of knee osteoarthritis grading into binary steps inside a semi-supervised teacher-student model with self-supervised reconstruction improves accuracy on limited labeled X-ray data.

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

The paper sets out to show that knee osteoarthritis severity grading can be made more reliable with scarce annotations by breaking the multi-class problem into a sequence of binary decisions. This is done inside a teacher-student semi-supervised architecture that also adds an adversarial reconstruction task to pull useful structure from unlabeled images and a feature-mixing step to sharpen grade boundaries when pseudo-labels are noisy. A reader would care because current methods need large expert-labeled sets and still falter on the natural imbalance between mild and severe cases, limiting their use in routine screening. If the approach holds, it points toward diagnostic tools that learn effectively from the unlabeled radiographs already stored in clinics.

Core claim

The authors claim that their H-SemiS framework, which fuses semi-supervised learning with self-supervision by decomposing severity grading into ordered binary sub-tasks, training an adversarial reconstruction module on unlabeled data, and applying quantum-inspired feature mixing inside the teacher-student loop, delivers higher performance than competing methods across accuracy, sensitivity, and other metrics on two multi-class knee X-ray datasets and generalizes to binary tasks.

What carries the argument

Hierarchical binary task decomposition inside a teacher-student semi-supervised architecture, combined with adversarial self-supervised reconstruction and quantum-inspired feature mixing to stabilize learning under class imbalance and noisy pseudo-labels.

Load-bearing premise

That the specific combination of binary decomposition, adversarial reconstruction, and quantum-inspired mixing will reliably lessen the damage from class imbalance and noisy labels without introducing its own biases on new clinical data.

What would settle it

Training and testing the full H-SemiS pipeline against a standard supervised baseline and a non-hierarchical semi-supervised version on a fresh, independent multi-center knee X-ray collection; if the proposed method no longer leads on the primary metrics, the central claim would not hold.

Figures

Figures reproduced from arXiv: 2604.23335 by Anushka Parwal, Chandravardhan Singh Raghaw, Nagendra Kumar, Prajakta Darade, Shahid Shafi Dar.

Figure 1
Figure 1. Figure 1: Diagnostic challenges in KXR analysis for KOA are highlighted from (a) to (f), including blurred pixels, indistinct joint space narrowing (JSN), low contrast, varying illumination, overlapping anatomical structures, and inconsistencies in scan composition (single or dual joint). 1.2. Existing Research and Gaps Several machine learning approaches have been explored for KOA diagnosis from KXR (Du et al., 201… view at source ↗
Figure 2
Figure 2. Figure 2: Gaps in existing research on severity grading of knee osteoarthritis are highlighted in the first row (i) to (iv). The second row (v) to (viii) presents the solutions offered by the proposed H-SemiS framework. 2 view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the proposed Hierarchical Semi-Supervised framework with Self-Supervision (H-SemiS) for knee osteoarthritis severity grading, consisting of three stages. Stage 1: Mask Image Reconstruction (MI-Rec) (refer Section 3.2) for KXR reconstruction; Stage 2: Similarity-aware Reconstructed Image Labeler (SiRL) (refer Section 3.3), assigns labels (KL-0 to KL-4) to reconstructed samples; and Stage 3: Hier… view at source ↗
Figure 4
Figure 4. Figure 4: Overview of our Masked Image Reconstruction (MI-Rec). The MI-Rec process begins with applying random masking to the input sample, dividing the image into visible and masked patches. Next, a reconstruction generator reconstructs the masked patches. Finally, the reconstruction discriminator evaluates the generated masked patches, classifying them as real or fake to produce the final unlabeled reconstructed s… view at source ↗
Figure 5
Figure 5. Figure 5: Overview of the Similarity-aware Reconstructed Image Labeler (SiRL) module for labeling unlabeled reconstructed KXR samples. The template library (top) is constructed using Wide Residual Networks (WRNs) to derive median feature vectors T m i for each KL grade. The similarity computation (bottom) assigns labels based on the highest similarity score. serves as the reference for subsequent similarity matching… view at source ↗
Figure 6
Figure 6. Figure 6: Overview of the Quantum-infused Teacher-Student (Q-TeSt) framework. Q-TeSt integrates convolutional networks with quantum convolutional networks in a teacher-student learning framework. The architecture leverages weakly and strongly augmented samples, transferring knowledge via EMA to optimize supervised and unsupervised learning objectives. 8 view at source ↗
Figure 7
Figure 7. Figure 7: Architectural overview of the Quantum Convolutional Network (QCN), comprising a quantum encoder, ansatz, and decoder. The quantum encoder processes and encodes the input data for transformation in the ansatz. The ansatz applies quantum gates to capture complex transformations through entanglement, and the decoder extracts the final outputs. 10 view at source ↗
Figure 8
Figure 8. Figure 8: Overview of the Hierarchical Multi-classifier (HiM) module for knee osteoarthritis severity grading. The HiM module operates in three stages. First, during the decomposition stage, the entire training dataset is divided based on the severity at the root node and sample distributions at subsequent nodes. Sec￾ond, in the classification stage, Q-TeSt models are trained at each level to grade osteoarthritis se… view at source ↗
Figure 9
Figure 9. Figure 9: Sample dataset used for evaluation of the proposed framework. The Multiclass dataset (left) includes five severity levels (KL-0 to KL-4) from the OAI and DKXI datasets. The Binary dataset (right) comprises normal (Class 0) and severe (Class 1) cases from the OP and KO datasets. Each severity level is color-coded for clarity. scans containing both knees; we crop the latter into individual knees for consiste… view at source ↗
Figure 10
Figure 10. Figure 10: Quantitative evaluation of the proposed H-SemiS framework (XIV) against existing baselines (I-XIII) for KOA severity grading. The radar plots compare the Acc, Pre, Rec, F1-score against different competing baselines on (a) OAI dataset (Chen, 2018) and (b) DKXI dataset (Gornale and Patravali, 2020). 14 view at source ↗
Figure 11
Figure 11. Figure 11: Confusion matrices displaying the classification performance on the OAI and DKXI datasets. puted as follows: χADCC = 3 view at source ↗
Figure 12
Figure 12. Figure 12: visualizes class activation maps with ADCC scores, demonstrating the ability of H-SemiS to localize clinically rel￾evant features. Higher ADCC values indicate more reliable and interpretable explanations aligned with the framework’s predic￾tions. A lower χAvD reflects reduced confidence when high￾lighted regions are removed, while χCo measures the consis￾tency of activation maps with expert interpretation… view at source ↗
Figure 13
Figure 13. Figure 13: KXR reconstruction examples across different KL grades. The first column represents the original images, while the subsequent columns display masked and reconstructed images with varying masking ratios (70%, 75%, and 80%). proposed MI-Rec. GAN-MAE achieves the second-best perfor￾mance, while MI-Rec consistently outperforms all alternatives across metrics. These results suggest that MI-Rec better pre￾serve… view at source ↗
Figure 14
Figure 14. Figure 14: Mask ratio ablation study highlighting the impact of masking on H￾SemiS performance. (ii) Effect of Template Matching view at source ↗
Figure 16
Figure 16. Figure 16: Impact of labeled data percentage on Teacher-Student framework ac￾curacy. Each quadrant shows performance at 1%, 5%, 10%, and 20% labeled data for three architectural configurations: Base Network (BN), BN with Quan￾tum Convolution Network (QCN), and BN with Normalization Layer (NL) and QCN. Quadrant XII (20% BN+NL+QCN) demonstrates the best performance. combined BN+NL+QCN configuration, demonstrating cons… view at source ↗
Figure 17
Figure 17. Figure 17: Performance comparison of H-SemiS under different decomposition techniques (No Decomposition, Random Decomposition, and Hierarchical Decompo￾sition) using original samples (OriS) and a combination of original and reconstructed samples (OriS+RecS). 18 view at source ↗
Figure 18
Figure 18. Figure 18: Performance comparison: ROC and PR curves for OP and KO datasets. 14.30 ms inference time, while Azizi et al. (2023) is more effi￾cient with 95.15M parameters and 5.85 ms inference. Among semi-supervised approaches, Nguyen et al. (2020) achieves the fastest inference at 3.22 ms with 25.35M parameters, whereas Huo et al. (2022) shows higher latency at 9.94 ms despite fewer parameters. In comparison, H-Semi… view at source ↗
Figure 19
Figure 19. Figure 19: Comparative analysis of model parameters, inference time, and accuracy for the H-SemiS framework against existing baselines on benchmark datasets. The first row illustrates the comparison between accuracy (x-axis) and the number of parameters (y-axis) on the OAI and DKXI datasets. The second row depicts the comparison between accuracy (x-axis) and inference time (y-axis) on the same datasets. 6.1. Feature… view at source ↗
Figure 20
Figure 20. Figure 20: The t-SNE visualization comparing features (a) without quantum enhancement and (b) with quantum enhancement. The plot of quantum-enhanced features effectively visualizes the multi-class classification, showing distinct and well-separated clusters for each class. tensive experiments on multiclass imbalanced datasets demon￾strate that H-SemiS consistently outperforms competing meth￾ods across key metrics. A… view at source ↗
read the original abstract

Knee osteoarthritis (KOA) is a degenerative joint disease that can lead to chronic pain, reduced mobility, and long-term disability. Automated severity grading from knee radiographs can support early assessment, but current methods heavily depend on large labeled datasets and remain sensitive to class imbalance, noisy samples, and variability in clinical annotations. To alleviate these limitations, we propose a Hierarchical fusion of Semi-Supervised framework with Self-Supervision (H-SemiS) for KOA severity grading in knee X-ray samples using limited annotated data. Rather than treating severity grading as a flat multi-class problem, H-SemiS decomposes the task into a sequence of binary sub-tasks within a semi-supervised teacher-student architecture, directly mitigating the impact of class imbalance. To further enhance feature learning from unlabeled data, the framework integrates an adversarial self-supervised reconstruction module that encourages the network to capture robust anatomical structures. In parallel, a teacher-student design with quantum-inspired feature mixing improves discrimination boundaries between adjacent grades when pseudo-labels are noisy. We comprehensively evaluate H-SemiS on two challenging multi-class datasets and assess its generalizability on two binary-class datasets. Our experimental results demonstrate the superiority of the proposed H-SemiS framework across multiple evaluation metrics, consistently outperforming several competing baselines and state-of-the-art methods. The code is publicly available at https://github.com/chandravardhan-singh-raghaw/H-SemiS.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 3 minor

Summary. The manuscript proposes H-SemiS, a hierarchical fusion of semi-supervised and self-supervised learning for knee osteoarthritis (KOA) severity grading from radiographs with limited labeled data. It decomposes the multi-class problem into sequential binary sub-tasks inside a teacher-student framework to mitigate class imbalance, adds an adversarial reconstruction module for robust feature learning from unlabeled samples, and introduces quantum-inspired feature mixing to sharpen boundaries between adjacent grades under noisy pseudo-labels. The method is evaluated on two multi-class KOA datasets and two binary datasets, with the central claim that it consistently outperforms competing baselines and state-of-the-art methods across multiple metrics.

Significance. If the empirical superiority holds after proper validation, the work could advance semi-supervised medical imaging by offering a structured decomposition that directly targets class imbalance and label noise in ordinal grading tasks like KOA. The public code release supports reproducibility, which is a positive factor for adoption in the field.

major comments (3)
  1. [§3.3] §3.3: The quantum-inspired feature mixing is described as improving discrimination under noisy pseudo-labels, yet no ablation is provided that isolates this operator from the hierarchical binary decomposition and teacher-student setup; without it, the load-bearing claim that the full framework mitigates imbalance and noise cannot be evaluated.
  2. [Experimental results section] Experimental results section (referenced in abstract and §4.2): The manuscript asserts consistent outperformance but supplies no statistical significance tests, confidence intervals, or error analysis on the metric gains; this directly weakens the central empirical claim.
  3. [§4.2] §4.2: No cross-dataset transfer experiments or analysis of decision boundaries between adjacent KL grades (e.g., KL-2 vs. KL-3) are reported, leaving open whether the reported gains generalize beyond the two specific KOA datasets or simply reflect dataset-specific tuning of the mixing weights.
minor comments (3)
  1. [Abstract] The abstract and introduction should explicitly name the two multi-class datasets and report their sizes and class distributions to allow immediate assessment of the experimental scope.
  2. [§3.3] Notation for the quantum-inspired mixing weights and the adversarial reconstruction loss weight should be introduced with explicit equations rather than descriptive text only.
  3. [Figure 1] Figure captions for the overall architecture should clarify the flow of labeled versus unlabeled data through the teacher-student and reconstruction modules.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback on our manuscript. We address each of the major comments point by point below, providing clarifications and committing to revisions where they will strengthen the work.

read point-by-point responses

  1. Referee: [§3.3] §3.3: The quantum-inspired feature mixing is described as improving discrimination under noisy pseudo-labels, yet no ablation is provided that isolates this operator from the hierarchical binary decomposition and teacher-student setup; without it, the load-bearing claim that the full framework mitigates imbalance and noise cannot be evaluated.

    Authors: We agree with the referee that isolating the contribution of the quantum-inspired feature mixing is important for validating its role in the framework. In the revised manuscript, we will add a dedicated ablation study that compares the full H-SemiS model against a variant without the quantum-inspired mixing, while keeping the hierarchical binary decomposition and teacher-student setup intact. This will allow readers to directly evaluate the impact of this component on handling noisy pseudo-labels and class imbalance. revision: yes

  2. Referee: [Experimental results section] Experimental results section (referenced in abstract and §4.2): The manuscript asserts consistent outperformance but supplies no statistical significance tests, confidence intervals, or error analysis on the metric gains; this directly weakens the central empirical claim.

    Authors: We acknowledge that the absence of statistical significance testing limits the strength of our empirical claims. We will revise the experimental results section to include appropriate statistical tests (such as paired t-tests across multiple runs) and report confidence intervals or standard deviations for the key metrics. This will provide a more robust quantification of the performance improvements over baselines. revision: yes

  3. Referee: [§4.2] §4.2: No cross-dataset transfer experiments or analysis of decision boundaries between adjacent KL grades (e.g., KL-2 vs. KL-3) are reported, leaving open whether the reported gains generalize beyond the two specific KOA datasets or simply reflect dataset-specific tuning of the mixing weights.

    Authors: Our current evaluation demonstrates consistent superiority across two multi-class KOA datasets with different characteristics and two binary datasets, which supports generalizability beyond a single dataset. However, we agree that explicit cross-dataset transfer experiments and focused analysis of decision boundaries between adjacent grades would further strengthen the claims. In the revision, we will include visualizations and quantitative analysis of decision boundaries (e.g., via confusion matrices or feature space projections for KL-2 vs. KL-3). We will also discuss the potential for cross-dataset transfer based on our multi-dataset results, though conducting full transfer experiments may be limited by dataset availability and will be noted as future work if not feasible within the revision timeline. revision: partial

Circularity Check

0 steps flagged

No circularity: purely empirical framework proposal with no derivations or self-referential predictions.

full rationale

The paper introduces the H-SemiS architecture (hierarchical binary decomposition in a teacher-student setup plus quantum-inspired feature mixing and adversarial self-supervised reconstruction) and reports its empirical superiority on two multi-class KOA datasets plus two binary datasets. No equations, first-principles derivations, fitted-parameter predictions, or uniqueness theorems appear in the provided text. Central claims rest on direct performance comparisons against external baselines rather than any reduction to inputs by construction. Self-citations, if present, are not load-bearing for any tautological step because no derivation chain exists to be circular.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 1 invented entities

Only the abstract is available; the ledger is therefore incomplete and based solely on stated design choices. The framework relies on standard deep-learning assumptions plus several domain-specific claims about the benefits of binary decomposition and quantum-inspired mixing.

free parameters (2)
  • hyperparameters of teacher-student training
    Learning rates, mixing coefficients, and pseudo-label thresholds are not specified in the abstract but are required for the semi-supervised pipeline.
  • adversarial reconstruction loss weight
    Balance between reconstruction and classification objectives is a free parameter typical in such hybrid models.
axioms (2)
  • domain assumption Decomposing severity grading into sequential binary sub-tasks directly mitigates class imbalance
    Invoked as the core motivation for the hierarchical design.
  • domain assumption Adversarial self-supervised reconstruction encourages capture of robust anatomical structures from unlabeled data
    Stated as the purpose of the self-supervision module.
invented entities (1)
  • quantum-inspired feature mixing no independent evidence
    purpose: Improves discrimination boundaries between adjacent grades when pseudo-labels are noisy
    Introduced as a novel component in the teacher-student design; no independent evidence provided.

pith-pipeline@v0.9.0 · 5581 in / 1467 out tokens · 72371 ms · 2026-05-08T08:42:29.005944+00:00 · methodology

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

55 extracted references · 55 canonical work pages

  1. [1]

    , author Culp, L

    author Azizi, S. , author Culp, L. , author Freyberg, J. , author Mustafa, B. , author Baur, S. , author Kornblith, S. , author Chen, T. , author Tomasev, N. , author Mitrović, J. , author Strachan, P. , author Mahdavi, S.S. , author Wulczyn, E. , author Babenko, B. , author Walker, M. , author Loh, A. , author Chen, P.H.C. , author Liu, Y. , author Bavis...

    work page doi:10.1038/s41551-023-01049-7 2023

  2. [2]

    In: 2021 IEEE/CVF International Conference on Computer Vision (ICCV)

    author Azizi, S. , author Mustafa, B. , author Ryan, F. , author Beaver, Z. , author Freyberg, J. , author Deaton, J. , author Loh, A. , author Karthikesalingam, A. , author Kornblith, S. , author Chen, T. , author Natarajan, V. , author Norouzi, M. , year 2021 . title Big self-supervised models advance medical image classification , in: booktitle 2021 IE...

    work page doi:10.1109/iccv48922.2021.00346 2021

  3. [3]

    Quantum Sci

    author Benedetti, M. , author Lloyd, E. , author Sack, S. , author Fiorentini, M. , year 2019 . title Parameterized quantum circuits as machine learning models . journal Quantum Science and Technology volume 4 , pages 043001 . :10.1088/2058-9565/ab4eb5

    work page doi:10.1088/2058-9565/ab4eb5 2019

  4. [4]

    , author Hans, D

    author Berrimi, M. , author Hans, D. , author Jennane, R. , year 2024 . title A semi-supervised multiview-mri network for the detection of knee osteoarthritis . journal Computerized Medical Imaging and Graphics volume 114 , pages 102371 . :10.1016/j.compmedimag.2024.102371

    work page doi:10.1016/j.compmedimag.2024.102371 2024

  5. [5]

    , author Myers, C

    author Burton, W. , author Myers, C. , author Rullkoetter, P. , year 2020 . title Semi-supervised learning for automatic segmentation of the knee from mri with convolutional neural networks . journal Computer Methods and Programs in Biomedicine volume 189 , pages 105328 . :10.1016/j.cmpb.2020.105328

    work page doi:10.1016/j.cmpb.2020.105328 2020

  6. [6]

    , author Chen, H

    author Cai, Y. , author Chen, H. , author Yang, X. , author Zhou, Y. , author Cheng, K.T. , year 2023 . title Dual-distribution discrepancy with self-supervised refinement for anomaly detection in medical images . journal Medical Image Analysis volume 86 , pages 102794 . :10.1016/j.media.2023.102794

    work page doi:10.1016/j.media.2023.102794 2023

  7. [7]

    , author Sarkar, A

    author Chattopadhay, A. , author Sarkar, A. , author Howlader, P. , author Balasubramanian, V.N. , year 2018 . title Grad-cam++: Generalized gradient-based visual explanations for deep convolutional networks , in: booktitle 2018 IEEE Winter Conference on Applications of Computer Vision (WACV) , publisher IEEE . pp. pages 839--847 . :10.1109/wacv.2018.00097

    work page doi:10.1109/wacv.2018.00097 2018

  8. [8]

    , year 2018

    author Chen, P. , year 2018 . title Knee osteoarthritis severity grading dataset . :0.17632/56rmx5bjcr.1

    work page 2018

  9. [9]

    , author Kornblith, S

    author Chen, T. , author Kornblith, S. , author Norouzi, M. , author Hinton, G. , year 2020 . title A simple framework for contrastive learning of visual representations , in: booktitle 37th International Conference on Machine Learning, ICML 2020 , pp. pages 1597--1607 . https://proceedings.mlr.press/v119/chen20j/chen20j.pdf. note accessed 8 March 2025

    work page 2020

  10. [10]

    Ancuti, C

    author Cubuk, E.D. , author Zoph, B. , author Shlens, J. , author Le, Q.V. , year 2020 . title Randaugment: Practical automated data augmentation with a reduced search space , in: booktitle 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) , publisher IEEE . pp. pages 702--703 . :10.1109/cvprw50498.2020.00359

    work page doi:10.1109/cvprw50498.2020.00359 2020

  11. [11]

    , author Beyer, L

    author Dosovitskiy, A. , author Beyer, L. , author Kolesnikov, A. , author Weissenborn, D. , author Zhai, X. , author Unterthiner, T. , author Dehghani, M. , author Minderer, M. , author Heigold, G. , author Gelly, S. , author Uszkoreit, J. , author Houlsby, N. , year 2021 . title An image is worth 16x16 words: Transformers for image recognition at scale ...

    work page 2021

  12. [12]

    , author Almajalid, R

    author Du, Y. , author Almajalid, R. , author Shan, J. , author Zhang, M. , year 2018 . title A novel method to predict knee osteoarthritis progression on mri using machine learning methods . journal IEEE Transactions on NanoBioscience volume 17 , pages 228–236 . :10.1109/tnb.2018.2840082

    work page doi:10.1109/tnb.2018.2840082 2018

  13. [13]

    , author Ullah, Z

    author Farooq, M.U. , author Ullah, Z. , author Khan, A. , author Gwak, J. , year 2023 . title Dc-aae: Dual channel adversarial autoencoder with multitask learning for kl-grade classification in knee radiographs . journal Computers in Biology and Medicine volume 167 , pages 107570 . :10.1016/j.compbiomed.2023.107570

    work page doi:10.1016/j.compbiomed.2023.107570 2023

  14. [14]

    In: 2023 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pp

    author Fei, Z. , author Fan, M. , author Zhu, L. , author Huang, J. , author Wei, X. , author Wei, X. , year 2023 . title Masked auto-encoders meet generative adversarial networks and beyond , in: booktitle 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , publisher IEEE . p. pages 24449–24459 . :10.1109/cvpr52729.2023.02342

    work page doi:10.1109/cvpr52729.2023.02342 2023

  15. [15]

    , author Kumar, S

    author Ghosh, S. , author Kumar, S. , author Kumar, A. , author Verma, J. , year 2024 . title A closer look at consistency regularization for semi-supervised learning , in: booktitle Proceedings of the 7th Joint International Conference on Data Science; Management of Data (11th ACM IKDD CODS and 29th COMAD) , publisher ACM . p. pages 10–17 . :10.1145/3632...

    work page doi:10.1145/3632410.3632453 2024

  16. [16]

    [HBP23] Aamal Abbas Hussain, Francesco Belardinelli, and G eorgios Piliouras

    author Goodfellow, I. , author Pouget-Abadie, J. , author Mirza, M. , author Xu, B. , author Warde-Farley, D. , author Ozair, S. , author Courville, A. , author Bengio, Y. , year 2020 . title Generative adversarial networks . journal Communications of the ACM volume 63 , pages 139–144 . :10.1145/3422622

    work page doi:10.1145/3422622 2020

  17. [17]

    , author Patravali, P

    author Gornale, S. , author Patravali, P. , year 2020 . title Digital knee x-ray images . :10.17632/T9NDX37V5H.1

    work page doi:10.17632/t9ndx37v5h.1 2020

  18. [18]

    In: IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022

    author Guo, J. , author Han, K. , author Wu, H. , author Tang, Y. , author Chen, X. , author Wang, Y. , author Xu, C. , year 2022 . title Cmt: Convolutional neural networks meet vision transformers , in: booktitle 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , publisher IEEE . p. pages 12165–12175 . :10.1109/cvpr52688.2022.01186

    work page doi:10.1109/cvpr52688.2022.01186 2022

  19. [19]

    In: IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022

    author He, K. , author Chen, X. , author Xie, S. , author Li, Y. , author Dollar, P. , author Girshick, R. , year 2022 . title Masked autoencoders are scalable vision learners , in: booktitle 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , publisher IEEE . pp. pages 16000--16009 . :10.1109/cvpr52688.2022.01553

    work page doi:10.1109/cvpr52688.2022.01553 2022

  20. [20]

    Deep residual learning for image recognition,

    author He, K. , author Zhang, X. , author Ren, S. , author Sun, J. , year 2016 . title Deep residual learning for image recognition , in: booktitle 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , publisher IEEE . pp. pages 770--778 . :10.1109/cvpr.2016.90

    work page doi:10.1109/cvpr.2016.90 2016

  21. [21]

    , author Wu, W

    author Hu, K. , author Wu, W. , author Li, W. , author Simic, M. , author Zomaya, A. , author Wang, Z. , year 2022 . title Adversarial evolving neural network for longitudinal knee osteoarthritis prediction . journal IEEE Transactions on Medical Imaging volume 41 , pages 3207–3217 . :10.1109/tmi.2022.3181060

    work page doi:10.1109/tmi.2022.3181060 2022

  22. [22]

    Densely connected convolutional networks,

    author Huang, G. , author Liu, Z. , author Van Der Maaten, L. , author Weinberger, K.Q. , year 2017 . title Densely connected convolutional networks , in: booktitle 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , publisher IEEE . pp. pages 4700--4708 . :10.1109/cvpr.2017.243

    work page doi:10.1109/cvpr.2017.243 2017

  23. [23]

    Medicine6, 74, DOI: 10.1038/s41746-023-00811-0 (2023)

    author Huang, S.C. , author Pareek, A. , author Jensen, M. , author Lungren, M.P. , author Yeung, S. , author Chaudhari, A.S. , year 2023 . title Self-supervised learning for medical image classification: a systematic review and implementation guidelines . journal npj Digital Medicine volume 6 . :10.1038/s41746-023-00811-0

    work page doi:10.1038/s41746-023-00811-0 2023

  24. [24]

    , author Jin, X

    author Huang, Z. , author Jin, X. , author Lu, C. , author Hou, Q. , author Cheng, M.M. , author Fu, D. , author Shen, X. , author Feng, J. , year 2024 . title Contrastive masked autoencoders are stronger vision learners . journal IEEE Transactions on Pattern Analysis and Machine Intelligence volume 46 , pages 2506–2517 . :10.1109/tpami.2023.3336525

    work page doi:10.1109/tpami.2023.3336525 2024

  25. [25]

    , author Ouyang, X

    author Huo, J. , author Ouyang, X. , author Si, L. , author Xuan, K. , author Wang, S. , author Yao, W. , author Liu, Y. , author Xu, J. , author Qian, D. , author Xue, Z. , author Wang, Q. , author Shen, D. , author Zhang, L. , year 2022 . title Automatic grading assessments for knee mri cartilage defects via self-ensembling semi-supervised learning with...

    work page doi:10.1016/j.media.2022.102508 2022

  26. [26]

    , author Lawrence, J

    author Kellgren, J. , author Lawrence, J. , year 1957 . title Radiological assessment of osteo-arthrosis . journal Annals of the Rheumatic Diseases volume 16 , pages 494–502 . :10.1136/ard.16.4.494

    work page doi:10.1136/ard.16.4.494 1957

  27. [27]

    , author Aila, T

    author Laine, S. , author Aila, T. , year 2017 . title Temporal ensembling for semi-supervised learning , in: booktitle 5th International Conference on Learning Representations, ICLR 2017 - Conference Track Proceedings , pp. pages 1--17 . https://openreview.net/forum?id=BJ6oOfqge. note accessed 8 March 2025

    work page 2017

  28. [28]

    , author Thiagarajah, S

    author Lindner, C. , author Thiagarajah, S. , author Wilkinson, J. , author Consortium, T. , author Wallis, G. , author Cootes, T. , year 2013 . title Fully automatic segmentation of the proximal femur using random forest regression voting . journal IEEE Transactions on Medical Imaging volume 32 , pages 1462–1472 . :10.1109/tmi.2013.2258030

    work page doi:10.1109/tmi.2013.2258030 2013

  29. [29]

    , author Lai, K.L

    author Lo, C.M. , author Lai, K.L. , year 2023 . title Deep learning-based assessment of knee septic arthritis using transformer features in sonographic modalities . journal Computer Methods and Programs in Biomedicine :10.1016/j.cmpb.2023.107575

    work page doi:10.1016/j.cmpb.2023.107575 2023

  30. [30]

    , author Hinton, G

    author van der Maaten, L. , author Hinton, G. , year 2008 . title Visualizing data using t-sne . journal Journal of Machine Learning Research volume 9 . https://jmlr.org/papers/volume9/vandermaaten08a/vandermaaten08a.pdf. note accessed 8 March 2025

    work page 2008

  31. [31]

    , author Ahmadabadi, H

    author Manzari, O.N. , author Ahmadabadi, H. , author Kashiani, H. , author Shokouhi, S.B. , author Ayatollahi, A. , year 2023 . title Medvit: A robust vision transformer for generalized medical image classification . journal Computers in Biology and Medicine volume 157 , pages 106791 . :10.1016/j.compbiomed.2023.106791

    work page doi:10.1016/j.compbiomed.2023.106791 2023

  32. [32]

    , author Bromley, T.R

    author Mari, A. , author Bromley, T.R. , author Izaac, J. , author Schuld, M. , author Killoran, N. , year 2020 . title Transfer learning in hybrid classical-quantum neural networks . journal Quantum volume 4 , pages 340 . :10.22331/q-2020-10-09-340

    work page doi:10.22331/q-2020-10-09-340 2020

  33. [33]

    , author Blaschko, M.B

    author Nguyen, H.H. , author Blaschko, M.B. , author Saarakkala, S. , author Tiulpin, A. , year 2024 . title Clinically-inspired multi-agent transformers for disease trajectory forecasting from multimodal data . journal IEEE Transactions on Medical Imaging volume 43 , pages 529–541 . :10.1109/tmi.2023.3312524

    work page doi:10.1109/tmi.2023.3312524 2024

  34. [34]

    , author Saarakkala, S

    author Nguyen, H.H. , author Saarakkala, S. , author Blaschko, M.B. , author Tiulpin, A. , year 2020 . title Semixup: In- and out-of-manifold regularization for deep semi-supervised knee osteoarthritis severity grading from plain radiographs . journal IEEE Transactions on Medical Imaging volume 39 , pages 4346–4356 . :10.1109/tmi.2020.3017007

    work page doi:10.1109/tmi.2020.3017007 2020

  35. [35]

    , author Jiang, Z

    author Ning, Z. , author Jiang, Z. , author Zhang, D. , year 2025 . title To combat multiclass imbalanced problems by aggregating evolutionary hierarchical classifiers . journal IEEE Transactions on Neural Networks and Learning Systems volume 36 , pages 5258–5272 . :10.1109/tnnls.2024.3383672

    work page doi:10.1109/tnnls.2024.3383672 2025

  36. [36]

    , author Avina-Cervantes, J.G

    author Ovalle-Magallanes, E. , author Avina-Cervantes, J.G. , author Cruz-Aceves, I. , author Ruiz-Pinales, J. , year 2022 . title Hybrid classical–quantum convolutional neural network for stenosis detection in x-ray coronary angiography . journal Expert Systems with Applications volume 189 , pages 116112 . :10.1016/j.eswa.2021.116112

    work page doi:10.1016/j.eswa.2021.116112 2022

  37. [37]

    , author Wu, Y

    author Pan, J. , author Wu, Y. , author Tang, Z. , author Sun, K. , author Li, M. , author Sun, J. , author Liu, J. , author Tian, J. , author Shen, B. , year 2024 . title Automatic knee osteoarthritis severity grading based on x-ray images using a hierarchical classification method . journal Arthritis Research & Therapy volume 26 . :10.1186/s13075-024-03416-4

    work page doi:10.1186/s13075-024-03416-4 2024

  38. [38]

    Jia, C., Yang, Y ., Xia, Y ., Chen, Y .-T., Parekh, Z., Pham, H., Le, Q., Sung, Y .-H., Li, Z., and Duerig, T

    author Poppi, S. , author Cornia, M. , author Baraldi, L. , author Cucchiara, R. , year 2021 . title Revisiting the evaluation of class activation mapping for explainability: A novel metric and experimental analysis , in: booktitle 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) , publisher IEEE . pp. pages 2299--2304...

    work page doi:10.1109/cvprw53098.2021.00260 2021

  39. [39]

    , author Li, H

    author Prabhakar, C. , author Li, H. , author Yang, J. , author Shit, S. , author Wiestler, B. , author Menze, B. , year 2024 . title Vit-ae++: improving vision transformer autoencoder for self-supervised medical image representations , in: booktitle Medical Imaging with Deep Learning , organization PMLR . pp. pages 666--679 . https://proceedings.mlr.pres...

    work page 2024

  40. [40]

    , author Bhore, P.S

    author Raghaw, C.S. , author Bhore, P.S. , author Rehman, M.Z.U. , author Kumar, N. , year 2024 . title An explainable contrastive-based dilated convolutional network with transformer for pediatric pneumonia detection . journal Applied Soft Computing volume 167 , pages 112258 . :10.1016/j.asoc.2024.112258

    work page doi:10.1016/j.asoc.2024.112258 2024

  41. [41]

    YOLO9000: Better, Faster, Stronger

    author Redmon, J. , author Farhadi, A. , year 2017 . title Yolo9000: Better, faster, stronger , in: booktitle 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , publisher IEEE . p. pages 6517–6525 . :10.1109/cvpr.2017.690

    work page doi:10.1109/cvpr.2017.690 2017

  42. [42]

    , author Srinivasan, C

    author Sam Chandra Bose, A. , author Srinivasan, C. , author Immaculate Joy, S. , year 2024 . title Optimized feature selection for enhanced accuracy in knee osteoarthritis detection and severity classification with machine learning . journal Biomedical Signal Processing and Control volume 97 , pages 106670 . :10.1016/j.bspc.2024.106670

    work page doi:10.1016/j.bspc.2024.106670 2024

  43. [43]

    doi:10.1007/s11263-019-01228-7

    author Selvaraju, R.R. , author Cogswell, M. , author Das, A. , author Vedantam, R. , author Parikh, D. , author Batra, D. , year 2019 . title Grad-cam: Visual explanations from deep networks via gradient-based localization . journal International Journal of Computer Vision volume 128 , pages 336–359 . :10.1007/s11263-019-01228-7

    work page doi:10.1007/s11263-019-01228-7 2019

  44. [44]

    , year 2024

    author Shaw, D. , year 2024 . title Osteoarthritis . :10.21227/mszn-gr21

    work page doi:10.21227/mszn-gr21 2024

  45. [45]

    , author Azad, M.M

    author Sohail, M. , author Azad, M.M. , author Kim, H.S. , year 2025 . title Knee osteoarthritis severity detection using deep inception transfer learning . journal Computers in Biology and Medicine volume 186 , pages 109641 . :10.1016/j.compbiomed.2024.109641

    work page doi:10.1016/j.compbiomed.2024.109641 2025

  46. [46]

    , author Du, P

    author Song, Z. , author Du, P. , author Yan, J. , author Li, K. , author Shou, J. , author Lai, M. , author Fan, Y. , author Xu, Y. , year 2024 . title Nucleus-aware self-supervised pretraining using unpaired image-to-image translation for histopathology images . journal IEEE Transactions on Medical Imaging volume 43 , pages 459–472 . :10.1109/tmi.2023.3309971

    work page doi:10.1109/tmi.2023.3309971 2024

  47. [47]

    , year 2023

    author Tao, M. , year 2023 . title Osteoarthritis prediction . :10.21227/EWX7-B315

    work page doi:10.21227/ewx7-b315 2023

  48. [48]

    , author Valpola, H

    author Tarvainen, A. , author Valpola, H. , year 2017 . title Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results , in: booktitle Proceedings of the 31st International Conference on Neural Information Processing Systems , p. pages 1195–1204

    work page 2017

  49. [49]

    , author Othmani, A

    author Teoh, Y.X. , author Othmani, A. , author Lai, K.W. , author Goh, S.L. , author Usman, J. , year 2023 . title Stratifying knee osteoarthritis features through multitask deep hybrid learning: Data from the osteoarthritis initiative . journal Computer Methods and Programs in Biomedicine volume 242 , pages 107807 . :10.1016/j.cmpb.2023.107807

    work page doi:10.1016/j.cmpb.2023.107807 2023

  50. [50]

    , author Cao, Y

    author Wang, F. , author Cao, Y. , author Lu, H. , author Pan, Y. , author Tao, Y. , author Huang, S. , author Wang, J. , author Huo, L. , author Wu, J. , year 2025 . title Osteoarthritis incidence trends globally, regionally, and nationally, 1990–2019: An age‐period‐cohort analysis . journal Musculoskeletal Care volume 23 . :10.1002/msc.70045

    work page doi:10.1002/msc.70045 2025

  51. [51]

    Ancuti, C

    author Wang, H. , author Wang, Z. , author Du, M. , author Yang, F. , author Zhang, Z. , author Ding, S. , author Mardziel, P. , author Hu, X. , year 2020 . title Score-cam: Score-weighted visual explanations for convolutional neural networks , in: booktitle 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) , publisher ...

    work page doi:10.1109/cvprw50498.2020.00020 2020

  52. [52]

    , author Song, D

    author Wang, Y. , author Song, D. , author Wang, W. , author Rao, S. , author Wang, X. , author Wang, M. , year 2022 . title Self-supervised learning and semi-supervised learning for multi-sequence medical image classification . journal Neurocomputing volume 513 , pages 383–394 . :10.1016/j.neucom.2022.09.097

    work page doi:10.1016/j.neucom.2022.09.097 2022

  53. [53]

    , author Hu, K

    author Wu, W. , author Hu, K. , author Yue, W. , author Li, W. , author Simic, M. , author Li, C. , author Xiang, W. , author Wang, Z. , year 2023 . title Self-supervised multimodal fusion network for knee osteoarthritis severity grading , in: booktitle 2023 International Conference on Digital Image Computing: Techniques and Applications (DICTA) , publish...

    work page doi:10.1109/dicta60407.2023.00017 2023

  54. [54]

    Wide residual networks

    author Zagoruyko, S. , author Komodakis, N. , year 2016 . title Wide residual networks , in: booktitle Procedings of the British Machine Vision Conference 2016 , publisher British Machine Vision Association . pp. pages 87.1--87.12 . :10.5244/c.30.87

    work page doi:10.5244/c.30.87 2016

  55. [55]

    , author Si, L

    author Zhuang, Z. , author Si, L. , author Wang, S. , author Xuan, K. , author Ouyang, X. , author Zhan, Y. , author Xue, Z. , author Zhang, L. , author Shen, D. , author Yao, W. , author Wang, Q. , year 2023 . title Knee cartilage defect assessment by graph representation and surface convolution . journal IEEE Transactions on Medical Imaging volume 42 , ...

    work page doi:10.1109/tmi.2022.3206042 2023