Pith. sign in

REVIEW 3 major objections 4 minor 77 references

UGPL: Uncertainty-Guided Progressive Learning for Evidence-Based Classification in Computed Tomography

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read UGPL classifies CT images by first scanning the whole image, then zooming into the regions it is least sure about; on three public CT datasets it reports accuracy gains of 3.29%, 2.46%, and 8.08% over state-of-the-art baselines.

desk verdict Solid system paper whose key ablation is undermined by a table-to-table inconsistency that looks like an honest but load-bearing error. read the letter →

arxiv 2507.14102 v1 pith:WWUM7MLU submitted 2025-07-18 eess.IV cs.CVcs.LG

classification eess.IVcs.CVcs.LG
keywords uncertainty-guidedprogressivelearningevidentialdeepcomputedtomographyclassificationadaptivefusionnon-maximumsuppressionkidneyabnormalitydetectionlungcancerCOVID-19
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper introduces UGPL, a framework for classifying computed tomography (CT) images that mimics a radiologist's two-pass reading: first a global pass produces a prediction and a pixel-wise uncertainty map, then a local pass crops the most uncertain regions and classifies them at higher resolution. The authors claim this uncertainty-guided progressive refinement consistently beats uniform-processing baselines, reporting 3.29% higher accuracy for kidney abnormalities, 2.46% for lung cancer type, and 8.08% for COVID-19 presence on three public CT datasets. They further claim that the uncertainty guidance itself is the key ingredient, with the full pipeline hugely outperforming variants that drop it. If these results hold, they suggest that directing computation to diagnostically ambiguous areas is a practical way to improve automated medical image diagnosis without simply scaling up the whole network.

What carries the argument

The central mechanism is the uncertainty-guided progressive patch-extraction loop. An evidential deep learning head produces a pixel-wise uncertainty map from a ResNet backbone; a greedy non-maximum suppression selects K patches centered on high-uncertainty locations while maintaining spatial diversity; a local refinement network classifies each patch and outputs a confidence score; an adaptive fusion MLP takes global logits plus mean global uncertainty and outputs a scalar weight that blends global and confidence-weighted local logits. This loop is what lets the model spend extra computation where the first pass is least certain.

What would settle it

Re-run the component ablation using the exact global model checkpoint from Table 2 as the global-only condition, and remeasure the full model's F1 gain on the same split; if the gain against that checkpoint is far below the reported 5.3x, the paper's central benefit is not supported.

Watch

Extended reading notes

Core claim

The central claim is that uncertainty—not attention or fixed multi-scale schedules—is the right signal for deciding where a classifier should look next. UGPL operationalizes this by training a global model whose evidential head outputs a spatial uncertainty map; a non-maximum-suppression patch extractor then crops the most uncertain regions, a local refinement network classifies each crop, and an adaptive fusion module weights global and local logits by confidence. Across kidney, lung cancer, and COVID-19 CT datasets, the fused model outperforms both branches alone and beats state-of-the-art baselines. The paper also reports that removing uncertainty guidance (random or fixed patches) or removing local refinement (global-only) sharply degrades performance, which the authors take as evidence that uncertainty-guided progressive allocation is what drives the gain.

Load-bearing premise

The claim that uncertainty guidance is what powers the gain rests on comparing the full model to a 'global-only' ablation whose reported accuracy (e.g., 0.2535 on COVID-19) is far below the accuracy the same global model supposedly achieves elsewhere in the paper (0.7108); if those two numbers are not produced by the same trained model, the baseline is unreliable.

Editorial extensions

If this is right

  • On three public CT datasets, the fused UGPL model reports accuracy of 99% (kidney), 98% (lung), and 81% (COVID-19), with accuracy gains of 3.29%, 2.46%, and 8.08% over the best baselines.
  • Uncertainty-guided patch selection beats both random and fixed patch locations in their ablations, and the full model outperforms its global-only, local-only, and unfused variants.
  • The best patch configuration is task-dependent: 64x64 patches with 3 crops for kidney, 2 for lung, and 4 for COVID-19, matching the spatial complexity of each disease.
  • Raising the uncertainty loss weight improves COVID-19 detection but slightly lowers kidney and lung performance, suggesting uncertainty calibration matters most when disease presentation is diffuse.

Reading between the lines

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

  • This suggests an active-learning extension: rank unlabeled CT slices by mean uncertainty and request labels for the most ambiguous ones, which fits with the paper's own future-work note.
  • Because fusion weight is a scalar learned per-image, a natural extension is per-patch fusion weights that could handle images with heterogeneous uncertainty across locations.
  • The same uncertainty map could drive a segmentation refinement stage, proposing regions for a segmentation network to re-examine—a neighbouring task the paper does not evaluate.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes UGPL, a two-stage framework for CT image classification: a global ResNet-based evidential model produces a spatial uncertainty map; a progressive patch extractor uses non-maximum suppression to select K high-uncertainty patches, which a local refinement network classifies; an adaptive fusion module combines the global and local logits. Training uses a weighted sum of seven losses. The experiments compare UGPL with thirteen baselines on kidney, lung-cancer, and COVID-19 CT datasets, reporting state-of-the-art accuracy and F1, and include component and ablation studies. The authors release code at an anonymous-style repository.

Significance. If the reported results are correct, the paper demonstrates a practical way to allocate computation to diagnostically uncertain spatial regions, and the release of code and the use of three public datasets are strengths. However, the central evidence for the uncertainty-guidance benefit — the ablation in Table 3 — is internally inconsistent with the component analysis in Table 2, so the magnitude of the claimed improvement (up to 5.3x F1) cannot currently be assessed. The contribution is therefore potentially valuable but not yet established.

major comments (3)
  1. [Table 3 vs. Table 2, Section 4.4.1] The ablation row labeled 'Global-only' in Table 3 is described exactly as 'the global uncertainty estimator without local refinement', which is the same component that Table 2 calls 'Global Model'. Yet the two tables report irreconcilable numbers: for COVID-19, Table 2 reports accuracy 0.7108 and F1 0.7078, while Table 3 reports 0.2535 and 0.1495; for lung cancer, 0.9617/0.9611 versus 0.5000/0.3890; for kidney, 0.9811/0.9746 versus 0.5676/0.5545. Because the full-model rows agree between the two tables (0.8108/0.7903 for COVID), the discrepancy is isolated to the ablation baselines. The conclusions 'upto 5.3x F1 improvement' (Section 5) and Figure 9 depend on the Table 3 baselines. Either Table 3 was produced with a different training protocol, or the global-only row reflects a broken or non-representative run (e.g., the random-selection fallback in Algorithm 1). The authors must reconcile these numbers or rerun the ablations; without a correct global-only baseline, the paper's central claim that uncertainty guidance is responsible for the gains is unsupported.
  2. [Section 4.4.1, Table 3] The text states that for kidney abnormalities 'the full model reaching 99.6% F1 versus 58.7% for the best ablated configuration (fixed patches)', but Table 3 lists the full-model kidney F1 as 0.9945 (99.45%) and the fixed-patches F1 as 0.5697 (56.97%). The quoted percentages are not the values in the table. This, together with the discrepancy in the previous comment, indicates that the ablation reporting is not reliable and must be corrected.
  3. [Abstract, Table 1] The abstract reports 'improvements of 3.29%, 2.46%, and 8.08% in accuracy' for kidney, lung, and COVID. These percentages cannot be derived from Table 1, where the best non-UGPL accuracies are 0.98 (CoaT), 0.95 (EfficientNetB0/ConvNeXt/CoaT), and 0.78 (DenseNet121), giving absolute/relative improvements of approximately 1.0%/1.0%, 3.0%/3.2%, and 3.0%/3.8%, respectively (or 0.81 minus 0.73 equals 0.08 for COVID against CRNet). The authors should state explicitly which baseline and which improvement measure (absolute percentage points, relative accuracy, etc.) these numbers refer to, and ensure they match Table 1.
minor comments (4)
  1. [Section 3.1, Eqs. (2)-(4); Supplementary A1.1.2] The pixel-wise uncertainty formula in Eq. (4) is introduced without derivation, and it is not apparent how the first term '1/alpha' equals aleatoric uncertainty or the second term equals epistemic uncertainty in the Dirichlet/subjective-logic framework of Eqs. (1)-(3); please provide a derivation or reference.
  2. [Figure 2 caption, Section 3.4] The figure caption lists loss abbreviations 'CE, UCC, CL, PDL, REG' while Section 3.4 defines CE, L_global, L_local, L_uncertainty, L_consistency, L_confidence, and L_diversity; please unify the notation.
  3. [Section 5] The text contains the typo 'upto' and the claim 'upto 5.3x F1 improvement' is unsupported until the ablation baselines are corrected.
  4. [Table 5] The table caption says 'Bolded values indicate results from C1 configuration', but no values are visibly bolded in the table; please format or clarify which entries are bolded.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central claim is an empirical benchmark comparison, and the ablation inconsistency is a data-validity concern rather than a definitional or self-citation reduction.

full rationale

UGPL's central claim is that uncertainty-guided progressive refinement outperforms baselines on three public CT datasets; this is an empirical result, not a derivation. The only internally load-bearing step is the uncertainty map's role in patch selection, and the uncertainty map is supervised by the uncertainty calibration loss Luncertainty = MSE(Uhat, 1 - C), where C is a correctness map derived from global predictions. Training an uncertainty estimator to flag errors and then exploiting those flags is ordinary supervised learning, not an output being derived from its own inputs. No parameter is fitted to the target metric and then re-reported as a prediction; the adaptive fusion weights are learned on training labels, and the final accuracy numbers come from held-out evaluation. The paper cites no prior work by its own authors and invokes no uniqueness theorem; EDL and subjective logic are standard external references. The reader's concern about Table 3's 'Global-only' row (COVID accuracy 0.2535, F1 0.1495) disagreeing with Table 2's 'Global Model' row (accuracy 0.7108, F1 0.7078) is substantial, but it is an internal consistency or experimental protocol problem, not circularity: the two rows are not defined as equal by any equation, and the reported 5.3x F1 improvement is arithmetic on the reported table entries. That issue belongs under correctness or reproducibility review, not under the circularity score.

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

The method is a deep learning pipeline with many learned weights, but the key free parameters beyond normal training are the patch count/size and the loss weights, both tuned on the validation results shown in Tables 4 and 5. The NMS margin, Gaussian shape, and diversity penalty are also hand-selected but unreported. The paper introduces no new physical or conceptual entities; the evidential parameters are internal network outputs. The main axioms are the validity of the bespoke uncertainty formula and the comparability of the ablation baselines.

free parameters (4)
  • Patch size P and patch count K per dataset = 64x64 and K=3 (kidney), K=2 (lung), K=4 (COVID)
    Chosen based on Table 5 as the best performing combination for each dataset; these choices affect the reported gains.
  • Loss weights (lambda_f, lambda_g, lambda_l, lambda_u, lambda_c, lambda_conf, lambda_d) = 1.0, 0.5, 0.5, 0.3, 0.2, 0.1, 0.1
    Baseline C1 selected from a 10-configuration sweep in Table 4; the paper states this is the best overall and uses it for the headline results.
  • NMS margin M and Gaussian suppression shape = not reported
    Algorithm 1 uses a margin M and Gaussian kernel G; values are not given, so these hand-selected parameters are not independently reproducible.
  • Diversity penalty lambda in patch selection (Eq. 6) = not reported
    The spatial penalty weight in the patch objective is not specified in the paper, although it controls patch diversity.
assumptions (4)
  • ad hoc to paper The pixel-wise uncertainty formula U in Eq. 4 correctly separates aleatoric and epistemic uncertainty.
    The formula is asserted as aleatoric plus epistemic without derivation; it is not the same as the EDL total uncertainty in Eq. 2, and the beta/nu parameterization is introduced specifically for this method.
  • domain assumption The global-only ablation in Table 3 is directly comparable to the global model in Table 2.
    The paper never states the training protocol difference, but the claim of a 5.3x F1 improvement depends on these baselines being valid.
  • domain assumption Pretrained ImageNet weights transfer to single-channel CT by averaging RGB channels.
    Section 3.2 and A1.1.1 modify the first conv layer in this way; if the pretrained features are not useful for CT, the method's performance would need re-tuning.
  • domain assumption The test labels and dataset splits for all three public datasets are correctly curated.
    The comparison is entirely based on these datasets; split details and fold counts are not provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of UGPL: Uncertainty-Guided Progressive Learning for Evidence-Based Classification in Computed Tomography." pith.science (2026). https://pith.science/paper/WWUM7MLU

@misc{pith2026250714102,
  author       = {Pith},
  title        = {Pith review of: UGPL: Uncertainty-Guided Progressive Learning for Evidence-Based Classification in Computed Tomography},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WWUM7MLU}},
  note         = {Machine review of arXiv:2507.14102}
}
read the original abstract

Accurate classification of computed tomography (CT) images is essential for diagnosis and treatment planning, but existing methods often struggle with the subtle and spatially diverse nature of pathological features. Current approaches typically process images uniformly, limiting their ability to detect localized abnormalities that require focused analysis. We introduce UGPL, an uncertainty-guided progressive learning framework that performs a global-to-local analysis by first identifying regions of diagnostic ambiguity and then conducting detailed examination of these critical areas. Our approach employs evidential deep learning to quantify predictive uncertainty, guiding the extraction of informative patches through a non-maximum suppression mechanism that maintains spatial diversity. This progressive refinement strategy, combined with an adaptive fusion mechanism, enables UGPL to integrate both contextual information and fine-grained details. Experiments across three CT datasets demonstrate that UGPL consistently outperforms state-of-the-art methods, achieving improvements of 3.29%, 2.46%, and 8.08% in accuracy for kidney abnormality, lung cancer, and COVID-19 detection, respectively. Our analysis shows that the uncertainty-guided component provides substantial benefits, with performance dramatically increasing when the full progressive learning pipeline is implemented. Our code is available at: https://github.com/shravan-18/UGPL

Figures

Figures reproduced from arXiv: 2507.14102 by the authors.

Figure 1
Figure 1. Comparison of medical image classification methods: [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The UGPL architecture pipeline. Our framework processes an input CT image through a global uncertainty estimator to produce [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Architecture of the Local Refinement Network. The network processes extracted patches [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Performance trends of model components across [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: ROC curves comparing global and fused model performance across datasets. The FM consistently maintains or improves the [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Comparison of t-SNE visualizations for feature spaces in the kidney dataset. (a) Global features from the full-image CNN form [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Uncertainty distribution by class for lung cancer de [PITH_FULL_IMAGE:figures/full_fig_p016_7.png]
Figure 10
Figure 10. Figure 10: Computational complexity (GFLOPs) versus inference [PITH_FULL_IMAGE:figures/full_fig_p018_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

77 extracted references · 73 canonical work pages

  1. [1]

    Rajendra Acharya, Ryszard Tadeusiewicz, and Saeid Nahavandi

    Moloud Abdar, Mohammad Amin Fahami, Satarupa Chakrabarti, Abbas Khosravi, Paweł Pławiak, U. Rajendra Acharya, Ryszard Tadeusiewicz, and Saeid Nahavandi. Barf: A new direct and cross-based binary residual feature fusion with uncertainty-aware module for medical image classifica- tion. Information Sciences, 577:353–378, 2021. 3

  2. [2]

    Frangi, U

    Moloud Abdar, Mohammad Amin Fahami, Leonardo Rundo, Petia Radeva, Alejandro F. Frangi, U. Rajendra Acharya, Abbas Khosravi, Hak-Keung Lam, Alexander Jung, and Saeid Nahavandi. Hercules: Deep hierarchical attentive multilevel fusion model with uncertainty quantifi- cation for medical image classification. IEEE Transactions on Industrial Informatics, 19(1):...

  3. [3]

    Al-Yasriy

    Muayed Al-Huseiny, Furat Mohsen, Enam Khalil, Zainab Hassan, Hamdalla Fadil, and Hamdalla F. Al-Yasriy. Eval- uation of svm performance in the detection of lung cancer in marked ct scan dataset. Indonesian Journal of Electrical Engineering and Computer Science, 21, 2021. 5, 6

  4. [4]

    The iq-othnccd lung cancer dataset,

    Hamdalla Alyasriy. The iq-othnccd lung cancer dataset,

  5. [5]

    Generating synthetic computed tomography (ct) im- ages to improve the performance of machine learning model for pediatric abdominal anomaly detection

    Samayan Bhattacharya, Avigyan Bhattacharya, and Sk Shah- nawaz. Generating synthetic computed tomography (ct) im- ages to improve the performance of machine learning model for pediatric abdominal anomaly detection. In Proceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 3865–3873, 2023. 1

  6. [6]

    Weight uncertainty in neural networks,

    Charles Blundell, Julien Cornebise, Koray Kavukcuoglu, and Daan Wierstra. Weight uncertainty in neural networks,

  7. [7]

    The diagnos- tic evaluation of convolutional neural network (cnn) for the assessment of chest x-ray of patients infected with covid-19

    Syed Usama Khalid Bukhari, Syed Safwan Khalid Bukhari, Asmara Syed, and Syed Sajid Hussain Shah. The diagnos- tic evaluation of convolutional neural network (cnn) for the assessment of chest x-ray of patients infected with covid-19. MedRxiv, pages 2020–03, 2020. 1

  8. [8]

    Iglovikov, Eugene Khved- chenya, Alex Parinov, Mikhail Druzhinin, and Alexandr A

    Alexander Buslaev, Vladimir I. Iglovikov, Eugene Khved- chenya, Alex Parinov, Mikhail Druzhinin, and Alexandr A. Kalinin. Albumentations: Fast and flexible image augmen- tations. Information, 11(2):125, 2020. 5

Show all 77 references
  1. [9]

    Paulo Chagas, Luiz Souza, Izabelle Pontes, Rodrigo Calumby, Michele Angelo, Angelo Duarte, Washington Lc- Dos Santos, and Luciano Oliveira. Uncertainty-aware mem- branous nephropathy classification: A monte-carlo dropout approach to detect how certain is the model.Computer Met...

  2. [10]

    CrossViT: Cross-Attention Multi-Scale Vision Trans- former for Image Classification

    Chun-Fu (Richard) Chen, Quanfu Fan, and Rameswar Panda. CrossViT: Cross-Attention Multi-Scale Vision Trans- former for Image Classification. In International Conference on Computer Vision (ICCV), 2021. 6

  3. [11]

    Recent advances and clin- ical applications of deep learning in medical image analysis

    Xuxin Chen, Ximin Wang, Ke Zhang, Kar-Ming Fung, Theresa C Thai, Kathleen Moore, Robert S Mannel, Hong Liu, Bin Zheng, and Yuchen Qiu. Recent advances and clin- ical applications of deep learning in medical image analysis. Medical image analysis, 79:102444, 2022. 1

  4. [12]

    Evil: Evidential inference learn- ing for trustworthy semi-supervised medical image segmen- tation

    Yingyu Chen, Ziyuan Yang, Chenyu Shen, Zhiwen Wang, Yang Qin, and Yi Zhang. Evil: Evidential inference learn- ing for trustworthy semi-supervised medical image segmen- tation. In 2023 IEEE 20th International Symposium on Biomedical Imaging (ISBI), pages 1–5, 2023. 3

  5. [13]

    Evidence-based uncertainty-aware semi- supervised medical image segmentation

    Yingyu Chen, Ziyuan Yang, Chenyu Shen, Zhiwen Wang, Zhongzhou Zhang, Yang Qin, Xin Wei, Jingfeng Lu, Yan Liu, and Yi Zhang. Evidence-based uncertainty-aware semi- supervised medical image segmentation. Computers in Biol- ogy and Medicine, 170:108004, 2024. 3

  6. [14]

    Pl-net: Progressive learning network for medical image segmentation, 2022

    Junlong Cheng, Chengrui Gao, Hongchun Lu, Zhangqiang Ming, Yong Yang, and Min Zhu. Pl-net: Progressive learning network for medical image segmentation, 2022. 2

  7. [15]

    A. P. Dempster. Upper and lower probabilities induced by a multivalued mapping. Annals of Mathematical Statistics, 38 (2):325–339, 1967. 3

  8. [16]

    An efficient model of residual based convo- lutional neural network with bayesian optimization for the classification of malarial cell images

    Aykut Diker. An efficient model of residual based convo- lutional neural network with bayesian optimization for the classification of malarial cell images. Computers in Biology and Medicine, 148:105635, 2022. 2

  9. [17]

    Dima et al

    A.F. Dima et al. 3d arterial segmentation via single 2d pro- jections and depth supervision in contrast-enhanced ct im- ages. In Medical Image Computing and Computer Assisted Intervention – MICCAI 2023. Springer, Cham, 2023. 1

  10. [18]

    Rconet: Deformable mutual information maximiza- tion and high-order uncertainty-aware learning for robust covid-19 detection

    Shunjie Dong, Qianqian Yang, Yu Fu, Mei Tian, and Cheng Zhuo. Rconet: Deformable mutual information maximiza- tion and high-order uncertainty-aware learning for robust covid-19 detection. IEEE Transactions on Neural Networks and Learning Systems, 32(8):3401–3411, 2021. 3

  11. [19]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition ...

  12. [20]

    Uncertainty quan- tification for deep unrolling-based computational imaging

    Canberk Ekmekci and Mujdat Cetin. Uncertainty quan- tification for deep unrolling-based computational imaging. IEEE Transactions on Computational Imaging , 8:1195– 1209, 2022. 3

  13. [21]

    Madu, Ismini Lourentzou, and Mehdi Moradi

    Mehrdad Eshraghi Dehaghani, Amirhossein Sabour, Ama- rachi B. Madu, Ismini Lourentzou, and Mehdi Moradi. Rep- resentation Learning with a Transformer-Based Detection Model for Localized Chest X-Ray Disease and Progression Detection . In proceedings of Medical Image Computing a...

  14. [22]

    Al-Yasriy, Muayed Al-Huseiny, Furat Mohsen, Enam Khalil, and Zainab Hassan

    Hamdalla F. Al-Yasriy, Muayed Al-Huseiny, Furat Mohsen, Enam Khalil, and Zainab Hassan. Diagnosis of lung cancer based on ct scans using cnn. IOP Conference Series: Mate- rials Science and Engineering, 928:022035, 2020. 5, 6

  15. [23]

    Evidence reconciled neural network for out-of-distribution detection in medical images

    Wei Fu, Yufei Chen, Wei Liu, Xiaodong Yue, and Chao Ma. Evidence reconciled neural network for out-of-distribution detection in medical images. In Medical Image Computing and Computer Assisted Intervention – MICCAI 2023 , pages 305–315, Cham, 2023. Springer Nature Switzerland. 3

  16. [24]

    Dropout as a bayesian approximation: Representing model uncertainty in deep learning, 2016

    Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: Representing model uncertainty in deep learning, 2016. 2, 3

  17. [25]

    Ghesu, Bogdan Georgescu, Awais Mansoor, Youngjin Yoo, Eli Gibson, R.S

    Florin C. Ghesu, Bogdan Georgescu, Awais Mansoor, Youngjin Yoo, Eli Gibson, R.S. Vishwanath, Abishek Bal- achandran, James M. Balter, Yue Cao, Ramandeep Singh, Subba R. Digumarthy, Mannudeep K. Kalra, Sasa Grbic, and Dorin Comaniciu. Quantifying and leveraging predic- tive unc...

  18. [26]

    A survey on attention mechanisms for medical applications: are we moving toward better algo- rithms? IEEE Access, 10:98909–98935, 2022

    Tiago Gonc ¸alves, Isabel Rio-Torto, Lu ´ıs F Teixeira, and Jaime S Cardoso. A survey on attention mechanisms for medical applications: are we moving toward better algo- rithms? IEEE Access, 10:98909–98935, 2022. 2

  19. [27]

    Distance-based detection of out-of-distribution silent failures for covid-19 lung lesion segmentation

    Camila Gonz ´alez, Karol Gotkowski, Moritz Fuchs, Andreas Bucher, Armin Dadras, Ricarda Fischbach, Isabel Jasmin Kaltenborn, and Anirban Mukhopadhyay. Distance-based detection of out-of-distribution silent failures for covid-19 lung lesion segmentation. Medical Image Analysis ...

  20. [28]

    Uncertainty-aware convo- lutional neural network for covid-19 x-ray images classifi- cation

    Mahesh Gour and Sweta Jain. Uncertainty-aware convo- lutional neural network for covid-19 x-ray images classifi- cation. Computers in Biology and Medicine , 140:105047,

  21. [29]

    Deep residual learning for image recognition, 2015

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition, 2015. 4, 6, 7, 1, 5

  22. [30]

    Supervised uncertainty quantifi- cation for segmentation with multiple annotations

    Shi Hu, Daniel Worrall, Stefan Knegt, Bas Veeling, Henkjan Huisman, and Max Welling. Supervised uncertainty quantifi- cation for segmentation with multiple annotations. In Medi- cal Image Computing and Computer Assisted Intervention – MICCAI 2019: 22nd International Conference...

  23. [31]

    Multi-view eviden- tial learning-based medical image segmentation

    Chao Huang, Yushu Shi, Waikeung Wong, Chengliang Liu, Wei Wang, Zhihua Wang, and Jie Wen. Multi-view eviden- tial learning-based medical image segmentation. Proceed- ings of the AAAI Conference on Artificial Intelligence , 39 (16):17386–17394, 2025. 3

  24. [32]

    Densely connected convolutional net- works

    Gao Huang, Zhuang Liu, Laurens Van Der Maaten, and Kil- ian Q Weinberger. Densely connected convolutional net- works. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4700–4708, 2017. 6, 7

  25. [33]

    Lymphoma segmentation from 3d pet-ct images using a deep evidential network

    Ling Huang, Su Ruan, Pierre Decazes, and Thierry Denœux. Lymphoma segmentation from 3d pet-ct images using a deep evidential network. International Journal of Approximate Reasoning, 149:39–60, 2022. 3

  26. [34]

    Deep evidential fusion with uncertainty quantification and reliability learning for multimodal medical image segmenta- tion

    Ling Huang, Su Ruan, Pierre Decazes, and Thierry Denœux. Deep evidential fusion with uncertainty quantification and reliability learning for multimodal medical image segmenta- tion. Information Fusion, 113:102648, 2025. 3

  27. [35]

    Qureshi, Jianqiang Li, and Tariq Mahmood

    Saeed Iqbal, Adnan N. Qureshi, Jianqiang Li, and Tariq Mahmood. On the analyses of medical images using tradi- tional machine learning techniques and convolutional neural networks. Archives of Computational Methods in Engineer- ing, 30:3173 – 3233, 2023. 1

  28. [36]

    M. N. Islam, M. Hasan, M. K. Hossain, et al. Vision trans- former and explainable transfer learning models for auto de- tection of kidney cyst, stone and tumor from ct-radiography. Scientific Reports, 12:11440, 2022. 5, 6

  29. [37]

    Principles of subjective networks

    Audun Jøsang and Lance Kaplan. Principles of subjective networks. In 2016 19th International Conference on Infor- mation Fusion (FUSION), pages 1292–1299, 2016. 4, 1

  30. [38]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 ,

  31. [39]

    An adaptive region- based transformer for nonrigid medical image registration with a self-constructing latent graph

    Sheng Lan, Xiu Li, and Zhenhua Guo. An adaptive region- based transformer for nonrigid medical image registration with a self-constructing latent graph. IEEE Transactions on Neural Networks and Learning Systems, 2023. 2

  32. [40]

    Drt: Deformable region-based transformer for nonrigid medical image regis- tration with a constraint of orientation

    Sheng Lan, Xiu Li, and Zhenhua Guo. Drt: Deformable region-based transformer for nonrigid medical image regis- tration with a constraint of orientation. IEEE Transactions on Instrumentation and Measurement, 72:1–15, 2023. 2

  33. [41]

    Region- based evidential deep learning to quantify uncertainty and improve robustness of brain tumor segmentation

    Hao Li, Yang Nan, Javier Del Ser, and Guang Yang. Region- based evidential deep learning to quantify uncertainty and improve robustness of brain tumor segmentation. Neural Comput. Appl., 35(30):22071–22085, 2022. 3

  34. [42]

    Swin transformer: Hierarchical vision transformer using shifted windows.2021 IEEE/CVF International Conference on Computer Vision (ICCV), pages 9992–10002, 2021

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows.2021 IEEE/CVF International Conference on Computer Vision (ICCV), pages 9992–10002, 2021. 6, 7

  35. [43]

    A convnet for the 2020s, 2022

    Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feicht- enhofer, Trevor Darrell, and Saining Xie. A convnet for the 2020s, 2022. 6

  36. [44]

    Sgdr: Stochastic gradient descent with warm restarts, 2017

    Ilya Loshchilov and Frank Hutter. Sgdr: Stochastic gradient descent with warm restarts, 2017. 6, 5

  37. [45]

    Trustworthy multimodal regression with mixture of normal-inverse gamma distribu- tions

    Huan Ma, Zongbo Han, Changqing Zhang, Huazhu Fu, Joey Tianyi Zhou, and Qinghua Hu. Trustworthy multimodal regression with mixture of normal-inverse gamma distribu- tions. In Proceedings of the 35th International Conference on Neural Information Processing Systems , Red Hook, N...

  38. [46]

    Shufflenet v2: Practical guidelines for efficient cnn architec- ture design

    Ningning Ma, Xiangyu Zhang, Hai-Tao Zheng, and Jian Sun. Shufflenet v2: Practical guidelines for efficient cnn architec- ture design. In Proceedings of the European Conference on Computer Vision (ECCV), pages 116–131, 2018. 6

  39. [47]

    Robinson, Bernhard Kainz, and Daniel Rueckert

    Qiang Ma, Liu Li, Emma C. Robinson, Bernhard Kainz, and Daniel Rueckert. Weakly Supervised Learning of Cortical Surface Reconstruction from Segmentations . Inproceedings of Medical Image Computing and Computer Assisted Inter- vention – MICCAI 2024. Springer Nature Switzerland, 2024. 1

  40. [48]

    Gastrointestinal abnormality detec- tion and classification using empirical wavelet transform and deep convolutional neural network from endoscopic images

    Subhashree Mohapatra, Girish Kumar Pati, Manohar Mishra, and Tripti Swarnkar. Gastrointestinal abnormality detec- tion and classification using empirical wavelet transform and deep convolutional neural network from endoscopic images. Ain Shams Engineering Journal, 14(4):101942...

  41. [49]

    Kaplano ˘glu, and A

    Nafiseh Ghaffar Nia, E. Kaplano ˘glu, and A. Nasab. Evalua- tion of artificial intelligence techniques in disease diagnosis and prediction. Discover Artificial Intelligence, 3, 2023. 1

  42. [50]

    Abnormality clas- sification and localization using dual-branch whole-region- based cnn model with histopathological images

    Olaide N Oyelade, Absalom E Ezugwu, Hein S Venter, Seyedali Mirjalili, and Amir H Gandomi. Abnormality clas- sification and localization using dual-branch whole-region- based cnn model with histopathological images. Computers in Biology and Medicine, 149:105943, 2022. 2

  43. [51]

    B. T. Polyak and A. B. Juditsky. Acceleration of stochastic approximation by averaging. SIAM Journal on Control and Optimization, 30(4):838–855, 1992. 5

  44. [52]

    Progres- sive generative adversarial network for generating high- dimensional and wide-frequency signals in intelligent fault diagnosis

    Zhijun Ren, Kai Huang, Yongsheng Zhu, Ke Feng, Zheng Liu, Hong Fu, Jun Hong, and Adam Glowacz. Progres- sive generative adversarial network for generating high- dimensional and wide-frequency signals in intelligent fault diagnosis. Engineering Applications of Artificial Intell...

  45. [53]

    Mobilenetv2: Inverted residuals and linear bottlenecks, 2019

    Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zh- moginov, and Liang-Chieh Chen. Mobilenetv2: Inverted residuals and linear bottlenecks, 2019. 6

  46. [54]

    Eviden- tial deep learning to quantify classification uncertainty, 2018

    Murat Sensoy, Lance Kaplan, and Melih Kandemir. Eviden- tial deep learning to quantify classification uncertainty, 2018. 3

  47. [55]

    Kebria, Darius Nahavandi, Saeid Nahavandi, and Dipti Srinivasan

    Afshar Shamsi, Hamzeh Asgharnezhad, Shirin Shamsi Jokandan, Abbas Khosravi, Parham M. Kebria, Darius Nahavandi, Saeid Nahavandi, and Dipti Srinivasan. An uncertainty-aware transfer learning-based framework for covid-19 diagnosis. IEEE Transactions on Neural Networks and Learni...

  48. [56]

    Dual-level deep ev- idential fusion: Integrating multimodal information for en- hanced reliable decision-making in deep learning

    Zhimin Shao, Weibei Dou, and Yu Pan. Dual-level deep ev- idential fusion: Integrating multimodal information for en- hanced reliable decision-making in deep learning. Informa- tion Fusion, 103:102113, 2024. 3

  49. [57]

    Very deep convo- lutional networks for large-scale image recognition

    Karen Simonyan and Andrew Zisserman. Very deep convo- lutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556, 2014. 6

  50. [58]

    Optimizing mcmc-driven bayesian neural networks for high-precision medical image classification in small sample sizes, 2024

    Mingyu Sun. Optimizing mcmc-driven bayesian neural networks for high-precision medical image classification in small sample sizes, 2024. 2

  51. [59]

    Deep evidential learning for radiotherapy dose prediction

    Hai Siong Tan, Kuancheng Wang, and Rafe McBeth. Deep evidential learning for radiotherapy dose prediction. Com- puters in Biology and Medicine, 182:109172, 2024. 3

  52. [60]

    Mingxing Tan and Quoc V . Le. Efficientnet: Rethinking model scaling for convolutional neural networks, 2020. 6, 7

  53. [61]

    X-ray and ct-scan- based automated detection and classification of covid-19 us- ing convolutional neural networks (cnn)

    Samritika Thakur and Aman Kumar. X-ray and ct-scan- based automated detection and classification of covid-19 us- ing convolutional neural networks (cnn). Biomedical Signal Processing and Control, 69:102920, 2021. 1

  54. [62]

    Training data-efficient image transformers & distillation through at- tention, 2021

    Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Herv ´e J´egou. Training data-efficient image transformers & distillation through at- tention, 2021. 6, 7

  55. [63]

    L. Wei, A. Yadav, and W. Hsu. Ctflow: Mitigating effects of computed tomography acquisition and reconstruction with normalizing flows. In Medical Image Computing and Com- puter Assisted Intervention – MICCAI 2023. Springer, Cham,

  56. [64]

    Alsaadi, and Nianyin Zeng

    Tingyi Xie, Zidong Wang, Han Li, Peishu Wu, Huixiang Huang, Hongyi Zhang, Fuad E. Alsaadi, and Nianyin Zeng. Progressive attention integration-based multi-scale efficient network for medical imaging analysis with application to covid-19 diagnosis. Computers in Biology and Medi...

  57. [65]

    Co- scale conv-attentional image transformers

    Weijian Xu, Yifan Xu, Tyler Chang, and Zhuowen Tu. Co- scale conv-attentional image transformers. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 9981–9990, 2021. 6

  58. [66]

    Isanet: Non-small cell lung cancer classification and detection based on cnn and attention mechanism

    Zhiwen Xu, Haijun Ren, Wei Zhou, and Zhichao Liu. Isanet: Non-small cell lung cancer classification and detection based on cnn and attention mechanism. Biomedical Signal Pro- cessing and Control, 77:103773, 2022. 2

  59. [67]

    WSSADN: A Weakly Supervised Spherical Age-Disentanglement Net- work for Detecting Developmental Disorders with Structural MRI

    Pengcheng Xue, Dong Nie, Meijiao Zhu, Ming Yang, Han Zhang, Daoqiang Zhang, and Xuyun Wen. WSSADN: A Weakly Supervised Spherical Age-Disentanglement Net- work for Detecting Developmental Disorders with Structural MRI . In proceedings of Medical Image Computing and Computer Ass...

  60. [68]

    Uncertainty quantification in medical im- age segmentation with multi-decoder u-net

    Yanwu Yang, Xutao Guo, Yiwei Pan, Pengcheng Shi, Haiyan Lv, and Ting Ma. Uncertainty quantification in medical im- age segmentation with multi-decoder u-net. In Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain In- juries: 7th International Workshop, BrainLes...

  61. [69]

    Brain ct image classification based on mask rcnn and atten- tion mechanism

    Shoulin Yin, Hang Li, Lin Teng, Asif Ali Laghari, Ahmad Almadhor, Michal Gregus, and Gabriel Avelino Sampedro. Brain ct image classification based on mask rcnn and atten- tion mechanism. Scientific Reports, 14(1):29300, 2024. 2

  62. [70]

    Three- way image classification with evidential deep convolutional neural networks

    Xiaodong Yue, Yufei Chen, Bin Yuan, and Ying Lv. Three- way image classification with evidential deep convolutional neural networks. Cognitive Computation, 14(6):2074–2086,

  63. [71]

    Monte-carlo frequency dropout for predic- tive uncertainty estimation in deep learning

    Tal Zeevi, Rajesh Venkataraman, Lawrence H Staib, and John A Onofrey. Monte-carlo frequency dropout for predic- tive uncertainty estimation in deep learning. In 2024 IEEE International Symposium on Biomedical Imaging (ISBI) , pages 1–5. IEEE, 2024. 2

  64. [72]

    An evidential-enhanced tri-branch consistency learning method for semi-supervised medical image segmentation

    Zhenxi Zhang, Heng Zhou, Xiaoran Shi, Ran Ran, Chunna Tian, and Feng Zhou. An evidential-enhanced tri-branch consistency learning method for semi-supervised medical image segmentation. IEEE Transactions on Instrumentation and Measurement, 73:1–13, 2024. 3

  65. [73]

    Evidence modeling for reliabil- ity learning and interpretable decision-making under multi- modality medical image segmentation

    Jianfeng Zhao and Shuo Li. Evidence modeling for reliabil- ity learning and interpretable decision-making under multi- modality medical image segmentation. Computerized Medi- cal Imaging and Graphics, 116:102422, 2024. 3

  66. [74]

    Covid-ct-dataset: a ct scan dataset about covid-19

    Jinyu Zhao, Yichen Zhang, Xuehai He, and Pengtao Xie. Covid-ct-dataset: a ct scan dataset about covid-19. arXiv preprint arXiv:2003.13865, 2020. 5, 6, 7

  67. [75]

    Ye Zhu, Jingwen Xu, Fei Lyu, and Pong C. Yuen. Symptom Disentanglement in Chest X-ray Images for Fine-Grained Progression Learning . In proceedings of Medical Image Computing and Computer Assisted Intervention – MICCAI

  68. [2021]

    OpenReview.net, 2021. 6, 7

  69. [2024]

    1 UGPL: Uncertainty-Guided Progressive Learning for Evidence-Based Classification in Computed Tomography Supplementary Material A1

    Springer Nature Switzerland, 2024. 1 UGPL: Uncertainty-Guided Progressive Learning for Evidence-Based Classification in Computed Tomography Supplementary Material A1. Extended Methodology A1.1. Global Uncertainty Estimation and Eviden- tial Learning Our global uncertainty esti...

Pith tools

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