Pith. sign in

REVIEW 4 major objections 6 minor 38 references

FOLC-Net: A Federated-Optimized Lightweight Architecture for Enhanced MRI Disease Diagnosis across Axial, Coronal, and Sagittal Views

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

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

Pith's one-line read FOLC-Net shows that a lightweight federated CNN can lift sagittal-view brain-tumor MRI accuracy to 92.44%, beating re-tested baselines by 2.33 to 4.07 percentage points while holding multi-view accuracy at 98.01%.

desk verdict Useful motivation and a sensible evaluation design, but the reported numbers don't add up; the internal test-set contradictions gut the central claim. read the letter →

arxiv 2507.06763 v1 pith:PQLTDGHJ submitted 2025-07-09 cs.CV cs.AI

Saif Ur Rehman Khan , Muhammad Nabeel Asim , Sebastian Vollmer , Andreas Dengel This is my paper · ORCID
classification cs.CVcs.AI
keywords federatedlearningbraintumorclassificationMRImulti-viewanalysislightweightCNNManta-rayforagingoptimizationConvNeXtintegrationsagittalviewperformancemedicalimagediagnosis
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 claims that the performance drop that state-of-the-art models show when MRI brain-tumor scans are evaluated on individual anatomical planes—axial, coronal, and especially sagittal—can be overcome by a federated, lightweight architecture called FOLC-Net. It reports that FOLC-Net reaches 92.44% accuracy on the sagittal view, compared with 88.37% to 90.11% for re-tested baselines, while keeping combined multi-view accuracy at 98.01% with only about 1.217 million parameters and 0.9 MB of storage. The authors argue that combining Manta-ray foraging optimization, global model cloning, and ConvNeXt-based client adaptation lets a shallow model adapt to view-specific features without sacrificing efficiency or privacy.

What carries the argument

The load-bearing mechanism is the MRFO-tuned ShallowFed CNN: a deliberately shallow convolutional network whose hyperparameters (filters, kernel sizes, activation, dropout, neurons) are optimized by the Manta-ray foraging optimizer's cyclone and chain foraging rules, then trained in a federated loop where the global model is cloned to each client, locally updated, and aggregated by data-proportional weights; ConvNeXt-style learning is integrated to improve client adaptability. This combination is what the paper credits for adapting to individual view characteristics rather than a single aggregated representation.

What would settle it

Retrain every baseline and FOLC-Net from scratch under a single shared protocol—same balanced augmentation, same 70/10/20 split, same 224x224 input, same optimizer settings, and the same test set—and check whether the sagittal-view accuracy gap (92.44% versus 88.37–90.11%) persists; if the margins shrink to noise or disappear, the central claim is an artifact of the comparison. A second check would be to run the identical FOLC-Net architecture without any MRFO tuning or ConvNeXt modification and compare per-view accuracies to isolate which component carries the improvement.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a shallow federated CNN, whose structure is tuned by Manta-ray foraging optimization and whose clients are empowered by ConvNeXt, can match or beat much larger networks on combined multi-view brain-tumor MRI classification and clearly outperform them on individual views. The largest reported gap is on the sagittal view, where FOLC-Net scores 92.44% accuracy versus 88.37% for the DL + Residual Learning baseline and 88.95% for the DL baseline, and the paper also reports gains on axial (99.44%) and coronal (98.27%) views. Because the model is evaluated per view, the paper argues, it exposes failure modes that combined-view evaluation hides and demonstrates a more clinically realistic picture of robustness.

Load-bearing premise

The comparison assumes the baselines in Tables 6 and 7 were re-trained and evaluated under the exact same data splits, preprocessing, augmentation, input size, and test set as FOLC-Net, yet the paper does not describe the baseline training protocol, so the reported margins over baselines could reflect evaluation setup rather than architectural superiority.

Editorial extensions

If this is right

  • If the central claim holds, multi-view medical imaging evaluations should report per-view metrics as a matter of course, since combined accuracy can hide substantial view-specific degradation.
  • Lightweight federated models of roughly 1.2 million parameters can plausibly be deployed on resource-constrained client devices such as edge hardware or hospital workstations without giving up accuracy.
  • The MRFO structure-search step offers a parameter-free way to adapt a small CNN to a new imaging modality, which the paper extends to CT and histopathology in its domain-shift tests.
  • The sagittal-view improvement specifically suggests that shallow, view-adapted models can handle the view where deep networks most often fail.

Reading between the lines

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

  • The reported margin may owe as much to the paper's class-balancing augmentation (for example, coronal and sagittal 'normal' classes are boosted from 34 and 78 training images to 483 and 495, respectively) as to the architecture; a controlled comparison training baselines on the same balanced data would isolate the effect.
  • 'Global model cloning' appears to be a weight-sharing scheme rather than a new aggregation algorithm, so the scalability claim is best tested by measuring communication rounds and convergence against standard FedAvg with the same architecture.
  • The MRFO search space is small and discrete; the resulting architecture resembles a conventional shallow CNN, which suggests the gains may come from the searched hyperparameters rather than from the federated loop itself—a claim that could be probed by training the same discovered architecture centrally.
  • A direct test of the privacy/utility trade-off—for instance, training FOLC-Net on data partitioned non-IID by view across clients versus centralized training—would clarify whether the federated setting or the architecture drives the performance.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes FOLC-Net, a federated learning framework that combines a lightweight CNN ('ShallowFed') with Manta-Ray Foraging Optimization (MRFO), global model cloning, and ConvNeXt blocks to improve brain-tumor MRI classification across axial, coronal, and sagittal views. The authors report state-of-the-art or near-state-of-the-art accuracy in all views (98.01% all-views, 99.44% axial, 98.27% coronal, 92.44% sagittal) and claim that the sagittal-view accuracy surpasses published baselines by about 2-4 percentage points. The paper also includes additional validation on an external binary MRI dataset, domain-shift tests on CT and histopathology data, interpretability analyses (Grad-CAM), t-SNE, and Chi-square tests. The central contribution is an architecture plus federated training recipe; the main evidence is empirical accuracy comparisons in Tables 5-7.

Significance. If the reported results were reproducible and the baselines were matched, the paper would be a useful contribution to federated medical image analysis: it targets a real problem (performance drop on individual anatomical views), demonstrates a lightweight model with a modest parameter count, and provides an unusually broad evaluation including interpretability and domain-shift checks. The work also follows reporting guidelines (TRIPOD), which is commendable. However, the paper does not ship code or data, does not report variance or seeds for the core accuracy claims, and contains several internal numerical inconsistencies that currently prevent the results from being verified. The significance of the contribution therefore cannot be assessed from the manuscript in its present form.

major comments (4)
  1. [Section 4.6, Table 3, Fig. 6] The confusion-matrix discussion contradicts the stated test-set sizes. Table 3 lists Sagittal Normal as having 10 test samples, but Section 4.6 states that the sagittal Normal class had 43 correct predictions and 5 misclassified (48 total). Similarly, Glioma sagittal has 45 test samples but the text reports 75 correct predictions, and Coronal Normal has 4 test samples but the text reports 50 correct predictions. These discrepancies are not explainable by augmentation or preprocessing because the test set is described as a held-out 20% split (Section 4.4). Since the per-view accuracies in Table 5 and the comparison in Table 7 rest on these confusion matrices, the headline sagittal accuracy of 92.44% cannot be reproduced from the paper's own evidence.
  2. [Section 4.11, Table 8] The reported Chi-square values exceed the theoretical maximum for the stated test-set sizes. For a 4x4 contingency table with N=172 (sagittal view), the maximum possible Chi-square value is N*(4-1)=516, but Table 8 reports 3500.4. Similarly, for the axial view (N=358, maximum 1074) the table reports 3800.2, and for the coronal view (N=174, maximum 522) it reports 3500.4. These values are impossible under the stated sample sizes, so the statistical validation as reported is invalid and cannot support the claim that predictions are strongly associated with true labels.
  3. [Sections 4.7-4.8, Tables 6-7] The baseline comparisons are not adequately matched. The paper says the baselines were 're-built and tested on each view separately' (Section 4.8), but it does not describe the training protocol for these baselines: optimizer, learning rate, epochs, batch size, augmentation, input resolution, or data-split seeds. Without this information and without error bars or repeated runs, the claimed margins (e.g., 92.44% vs 90.11% in the sagittal view) could be artifacts of evaluation setup or run-to-run variation, particularly given the small sagittal test set of only 172 images. The manuscript should provide code, seeds, per-run results, and a precise description of the baseline training setup to make the comparison meaningful.
  4. [Section 3.1.3-3.1.4] The MRFO architecture search selects the best ShallowFed structure by maximizing validation accuracy on the same dataset that is later used for reporting test performance. This creates a selection bias on the validation set that is not accounted for in the reported numbers. The paper should either describe a nested evaluation (e.g., separate architecture-search and final test splits) or discuss the potential optimism in the reported accuracies resulting from validation-based model selection.
minor comments (6)
  1. [Section 4.12] The parameter counts are stated as '2,710,000 million parameters' and '1,217,000 million parameters', which are arithmetically 2.71 billion and 1.217 billion, respectively; the abstract says 1.217 million. These should be corrected to 'approximately 2.71 million' and 'approximately 1.217 million' or to the exact values.
  2. [Section 3.1.3] 'MFRO' appears as a misspelling of 'MRFO' in several places; please correct this.
  3. [Section 4.6] The method names 'GRADCAN' and 'GRADCAN++' should be 'Grad-CAM' and 'Grad-CAM++' throughout.
  4. [Tables 3 and 4] Table 4 appears to show augmented/balanced training counts while Table 3 shows raw counts; the caption or text should explicitly state that Table 4 reflects data balancing/augmentation to avoid an apparent contradiction.
  5. [Declarations, Table 2] The Data Availability statement says 'Data will be made available on request,' yet the TRIPOD checklist in Table 2 marks Data and Code Availability as 'Unclear.' The authors should clarify whether they will release code and exact data splits for reproducibility.
  6. [Section 4.6, Fig. 6] The text describing the axial confusion matrix claims that the Pituitary class has 'the most significant number of misclassifications (1 misclassified),' but the reported 99.44% axial accuracy requires more than one misclassification if the test set has 358 samples; this should be reconciled.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reported accuracies are empirical test-set results, not derivations from their own inputs; self-citations and internal numerical inconsistencies are reproducibility concerns, not circular reasoning.

full rationale

The central claim that FOLC-Net attains 98.01% all-view accuracy and 92.44% sagittal-view accuracy is an empirical evaluation reported in Tables 5-7 and Section 4.8, not a derivation from fitted parameters or from prior results. The MRFO structure search in Section 3.1.4 maximizes validation accuracy (Eq. 11), and Section 4.4 states that a separate 20% per-class test split was held out; reporting the held-out test accuracy is therefore not a prediction of the fitted objective. The paper cites several works by the same authors ([10], [29], [30]) as baselines, and Table 7 says these models were re-built and tested on each view, so this is a conventional empirical comparison rather than a self-citation chain that replaces evidence. Concerns about missing baseline training details, conflicting sample counts (Table 3 vs. Section 4.6), and the paper's own 'Unclear' TRIPOD entries (Table 2) are legitimate correctness or reproducibility issues, but they do not make a quantity equivalent to its own inputs. No equation or fitted parameter has been shown to reduce to its own output, so no circularity step meets the evidentiary bar.

Assumptions & free parameters 3 free parameters · 3 assumptions · 2 invented entities

The central claim rests on the architecture chosen by MRFO (a fitting procedure), on an undocumented baseline comparison, and on the reported split of a public dataset. The design choices of search space, epochs, batch size, optimizer, and client configuration are all hand-set inputs that are not independently justified.

free parameters (3)
  • MRFO search space bounds = filters 8-512, kernels 3x3-9x9, activations ReLU/LeakyReLU/Tanh/ELU, dropout 0.1-0.5, neurons 16-128
    The final architecture is selected by MRFO maximizing validation accuracy on the same dataset. The search space boundaries (Table 1) are chosen by the authors and the resulting architecture (32 filters in first conv, LeakyReLU alpha 0.1, dropout 0.25, dense 128/64, dropout 0.5) is a fitted configuration, not a derivation.
  • MRFO population and iteration settings = population 10, 8 cycles of 40 epochs, patience 10
    Section 3.1.3 states these values without sensitivity analysis. The final model depends on them and they are hand-set.
  • Global cloning and client count = N = C5 (undefined)
    Section 3.5 says 'In our case, it is C5' but the number of clients and data partition per client are never defined. The cloned global model and weighted aggregation depend on this unspecified structure.
assumptions (3)
  • domain assumption Validation accuracy is a suitable objective for architecture search (Eq. 11).
    The fitness function maximizes validation accuracy on the same Kaggle dataset; the final test set is held out, but no cross-validation or repeated search is reported, so the search may overfit to the validation split.
  • domain assumption The baseline models were evaluated under the same data split, preprocessing, and augmentation as FOLC-Net.
    Tables 6-7 claim superior performance over DenseNet/ResNet/VGG/MobileNet and over the cited methods, but the text gives no protocol for the baseline training. This is a load-bearing assumption for every comparison claim.
  • domain assumption Federated averaging with weight scaling (Eq. 12-17) produces a single global model that can be evaluated on the centralized test set.
    The evaluation pipeline aggregates client models and then reports metrics on a held-out test set. The paper does not compare against a centralized non-federated training baseline with the same architecture, so the marginal contribution of the federated protocol is not isolated.
invented entities (2)
  • ShallowFed
    purpose: A lightweight CNN architecture that serves as the global and client model.
    The architecture is described textually (conv blocks, pooling, dropout, dense layers) and is claimed to be novel, but no standalone evidence or code is provided beyond the reported accuracy tables.
  • Global model cloning
    purpose: Replicating the global model at each client to reduce communication and training cost.
    This is presented as a novel mechanism, but the description (Section 3.4) matches standard federated learning where clients start from the global model. No ablation isolates the effect of cloning.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FOLC-Net: A Federated-Optimized Lightweight Architecture for Enhanced MRI Disease Diagnosis across Axial, Coronal, and Sagittal Views." pith.science (2026). https://pith.science/paper/PQLTDGHJ

@misc{pith2026250706763,
  author       = {Pith},
  title        = {Pith review of: FOLC-Net: A Federated-Optimized Lightweight Architecture for Enhanced MRI Disease Diagnosis across Axial, Coronal, and Sagittal Views},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PQLTDGHJ}},
  note         = {Machine review of arXiv:2507.06763}
}
read the original abstract

The framework is designed to improve performance in the analysis of combined as well as single anatomical perspectives for MRI disease diagnosis. It specifically addresses the performance degradation observed in state-of-the-art (SOTA) models, particularly when processing axial, coronal, and sagittal anatomical planes. The paper introduces the FOLC-Net framework, which incorporates a novel federated-optimized lightweight architecture with approximately 1.217 million parameters and a storage requirement of only 0.9 MB. FOLC-Net integrates Manta-ray foraging optimization (MRFO) mechanisms for efficient model structure generation, global model cloning for scalable training, and ConvNeXt for enhanced client adaptability. The model was evaluated on combined multi-view data as well as individual views, such as axial, coronal, and sagittal, to assess its robustness in various medical imaging scenarios. Moreover, FOLC-Net tests a ShallowFed model on different data to evaluate its ability to generalize beyond the training dataset. The results show that FOLC-Net outperforms existing models, particularly in the challenging sagittal view. For instance, FOLC-Net achieved an accuracy of 92.44% on the sagittal view, significantly higher than the 88.37% accuracy of study method (DL + Residual Learning) and 88.95% of DL models. Additionally, FOLC-Net demonstrated improved accuracy across all individual views, providing a more reliable and robust solution for medical image analysis in decentralized environments. FOLC-Net addresses the limitations of existing SOTA models by providing a framework that ensures better adaptability to individual views while maintaining strong performance in multi-view settings. The incorporation of MRFO, global model cloning, and ConvNeXt ensures that FOLC-Net performs better in real-world medical applications.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 37 canonical work pages

  1. [1]

    Journal of Pharmacognosy and Phytochemistry 14(1), 272–275 (2025)

    Sharma, A., et al.: Cancer: Pathology of the century act as big killers: A disease with uncontrolled division of cells. Journal of Pharmacognosy and Phytochemistry 14(1), 272–275 (2025)

    work page 2025

  2. [2]

    Journal of Imaging 7(9), 179 (2021)

    Biratu, E.S., et al.: A survey of brain tumor segmentation and classification algorithms. Journal of Imaging 7(9), 179 (2021)

    work page 2021

  3. [3]

    European Journal of Medical Research 28(1), 39 (2023)

    Salari, N., et al.: The global prevalence of primary central nervous system tumors: a systematic review and meta-analysis. European Journal of Medical Research 28(1), 39 (2023)

    work page 2023

  4. [4]

    Brain & Neurorehabilitation 15(3), 25 (2022)

    Park, J., Park, Y.G.: Brain tumor rehabilitation: symptoms, complications, and treatment strategy. Brain & Neurorehabilitation 15(3), 25 (2022)

    work page 2022

  5. [5]

    Archives of Computational Methods in Engineering 29(7), 4871–4896 (2022)

    Ali, S., et al.: A comprehensive survey on brain tumor diagnosis using deep learning and emerg- ing hybrid techniques with multi-modal mr image. Archives of Computational Methods in Engineering 29(7), 4871–4896 (2022)

    work page 2022

  6. [6]

    Stroke 51(8), 2573–2579 (2020)

    Mouridsen, K., Thurner, P., Zaharchuk, G.: Artificial intelligence applications in stroke. Stroke 51(8), 2573–2579 (2020)

    work page 2020

  7. [7]

    Applied Soft Computing 167, 112366 (2024)

    Bilal, O., et al.: Differential evolution optimization based ensemble framework for accurate cervical cancer diagnosis. Applied Soft Computing 167, 112366 (2024)

    work page 2024

  8. [8]

    International Journal of Imaging Systems and Technology 35(3), 70090 (2025)

    Khan, S.U.R., Asif, S., Bilal, O.: Ensemble architecture of vision transformer and cnns for breast cancer tumor detection from mammograms. International Journal of Imaging Systems and Technology 35(3), 70090 (2025)

    work page 2025

Show all 38 references
  1. [9]

    Computers and Electrical Engineering 123, 110106 (2025) 36

    Bilal, O., et al.: An amalgamation of deep neural networks optimized with salp swarm algorithm for cervical cancer detection. Computers and Electrical Engineering 123, 110106 (2025) 36

    2025

  2. [10]

    Journal of Bionic Engineering, 1–20 (2025)

    Khan, S.U.R., Khan, Z.: Detection of abnormal cardiac rhythms using feature fusion technique with heart sound spectrograms. Journal of Bionic Engineering, 1–20 (2025)

    2025

  3. [11]

    International Journal of Machine Learning and Cybernetics, 1–20 (2025)

    Khan, S.U.R., et al.: Lead-cnn: lightweight enhanced dimension reduction convolutional neu- ral network for brain tumor classification. International Journal of Machine Learning and Cybernetics, 1–20 (2025)

    2025

  4. [12]

    Biomedical Signal Processing and Control 109, 108040 (2025)

    Hekmat, A., et al.: Brain tumor diagnosis redefined: Leveraging image fusion for mri enhance- ment classification. Biomedical Signal Processing and Control 109, 108040 (2025)

    2025

  5. [13]

    Biomedical Signal Processing and Control 76, 103631 (2022)

    Kumar, K.A., Prasad, A., Metan, J.: A hybrid deep cnn-cov-19-res-net transfer learning architype for an enhanced brain tumor detection and classification scheme in medical image processing. Biomedical Signal Processing and Control 76, 103631 (2022)

    2022

  6. [14]

    In: International MICCAI Brainlesion Workshop

    Chowdhury, A., et al.: A review of medical federated learning: Applications in oncology and cancer research. In: International MICCAI Brainlesion Workshop. Springer, ??? (2021)

    2021

  7. [15]

    Pattern Recognition, 110424 (2024)

    Guan, H., et al.: Federated learning for medical image analysis: A survey. Pattern Recognition, 110424 (2024)

    2024

  8. [16]

    Bulletin of Electrical Engineering and Informatics 11(6), 3313–3321 (2022)

    Mahlool, D.H., Abed, M.H.: Distributed brain tumor diagnosis using a federated learning environment. Bulletin of Electrical Engineering and Informatics 11(6), 3313–3321 (2022)

    2022

  9. [17]

    https://arxiv.org/abs/2507.04249

    Khan, U.S., Khan, S.U.R.: Ethics by Design: A Lifecycle Framework for Trustworthy AI in Medical Imaging From Transparent Data Governance to Clinically Validated Deployment (2025). https://arxiv.org/abs/2507.04249

    2025 arXiv

  10. [18]

    BMC Medical Informatics and Decision Making 23(1), 16 (2023)

    Saeedi, S., et al.: Mri-based brain tumor detection using convolutional deep learning methods and chosen machine learning techniques. BMC Medical Informatics and Decision Making 23(1), 16 (2023)

    2023

  11. [19]

    Multimedia Tools and Applications 83(7), 20487–20506 (2024)

    Rajput, I.S., et al.: A transfer learning-based brain tumor classification using magnetic resonance images. Multimedia Tools and Applications 83(7), 20487–20506 (2024)

    2024

  12. [20]

    Journal of Computational Science 30, 174–182 (2019)

    Sajjad, M., et al.: Multi-grade brain tumor classification using deep cnn with extensive data augmentation. Journal of Computational Science 30, 174–182 (2019)

    2019

  13. [21]

    Neural Processing Letters 55(4), 3779–3809 (2023)

    Islam, M., et al.: Effectiveness of federated learning and cnn ensemble architectures for identifying brain tumors using mri images. Neural Processing Letters 55(4), 3779–3809 (2023)

    2023

  14. [22]

    Frontiers in Oncology15, 1535478 (2025)

    Mastoi, Q.-U.-A., et al.: Explainable ai in medical imaging: an interpretable and collaborative federated learning model for brain tumor classification. Frontiers in Oncology15, 1535478 (2025)

    2025

  15. [23]

    IEEE Access (2025)

    Sivakumar, N., et al.: A hybrid brain tumor classification using fl with fedavg and fedprox for privacy and robustness across heterogeneous data sources. IEEE Access (2025)

    2025

  16. [24]

    In: 2024 IEEE Region 10 Symposium (TENSYMP)

    Rajit, S., et al.: Multi-class brain tumor classification of mri image using federated learning with blockchain. In: 2024 IEEE Region 10 Symposium (TENSYMP). IEEE, ??? (2024)

    2024

  17. [25]

    Ing´ enierie des Syst` emes d’Information 30(4) (2025)

    Jewalikar, A., Khan, R.A.H., Mane, D.T.: Efficient federated aggregation technique for brain tumor classification using transfer learning approaches. Ing´ enierie des Syst` emes d’Information 30(4) (2025)

    2025

  18. [26]

    In: KDD 2024 Workshop-Artificial Intelligence and Data Science for Healthcare, Barcelona, Spain, pp

    Lhasnaoui, C., et al.: Decentralized diagnosis: Privacy-preserving brain tumor classification with federated learning. In: KDD 2024 Workshop-Artificial Intelligence and Data Science for Healthcare, Barcelona, Spain, pp. 25–29 (2024)

    2024

  19. [27]

    In: 2024 2nd International Conference on Information and Communication Technology (ICICT)

    Samsuzzaman, M., et al.: Optimizing privacy and efficiency in brain tumor classification through advanced non-iid federated deep learning. In: 2024 2nd International Conference on Information and Communication Technology (ICICT). IEEE, ??? (2024) 37

    2024

  20. [28]

    Multimedia Systems 31(1), 1–27 (2025)

    Khan, S.U.R., et al.: Optimize brain tumor multiclass classification with manta ray foraging and improved residual block techniques. Multimedia Systems 31(1), 1–27 (2025)

    2025

  21. [29]

    In: 2024 International Conference on Digital Image Computing: Techniques and Applications (DICTA)

    Khan, Z., et al.: Boosting the prediction of brain tumor using two stage bigait architecture. In: 2024 International Conference on Digital Image Computing: Techniques and Applications (DICTA). IEEE, ??? (2024)

    2024

  22. [30]

    International Journal of Imaging Systems and Technology 34(1), 22975 (2024)

    Khan, S.U.R., et al.: Hybrid-net: A fusion of densenet169 and advanced machine learning classifiers for enhanced brain tumor diagnosis. International Journal of Imaging Systems and Technology 34(1), 22975 (2024)

    2024

  23. [31]

    Journal of King Saud University- Computer and Information Sciences 35(8), 101665 (2023)

    Saleem, S., Asim, M.N., Van Elst, L., Dengel, A.: Fnreq-net: A hybrid computational framework for functional and non-functional requirements classification. Journal of King Saud University- Computer and Information Sciences 35(8), 101665 (2023)

    2023

  24. [32]

    International Journal of Molecular Sciences 23(15), 8221 (2022)

    Asim, M.N., Ibrahim, M.A., Imran Malik, M., Dengel, A., Ahmed, S.: Circ-locnet: A compu- tational framework for circular rna sub-cellular localization prediction. International Journal of Molecular Sciences 23(15), 8221 (2022)

    2022

  25. [33]

    In: 2020 Fourth International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud)(I- SMAC), pp

    Siddique, M.A.B., Sakib, S., Khan, M.M.R., Tanzeem, A.K., Chowdhury, M., Yasmin, N.: Deep convolutional neural networks model-based brain tumor detection in brain mri images. In: 2020 Fourth International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud)(I- SMA...

    2020

  26. [34]

    Multimedia Tools and Applications, 1–28 (2025)

    Khan, M.A., Park, H.: Adaptive channel attention and multi-path convolutional architecture for brain tumor detection using mri images. Multimedia Tools and Applications, 1–28 (2025)

    2025

  27. [35]

    Concurrency and Computation: Practice and Experience 34(1), 6541 (2022)

    Isunuri, B.V., Kakarla, J.: Three-class brain tumor classification from magnetic resonance images using separable convolution based neural network. Concurrency and Computation: Practice and Experience 34(1), 6541 (2022)

    2022

  28. [36]

    International Journal of Intelligent Networks 3, 188–196 (2022)

    Shanthi, S., et al.: An efficient automatic brain tumor classification using optimized hybrid deep neural network. International Journal of Intelligent Networks 3, 188–196 (2022)

    2022

  29. [37]

    Computers, Materials & Continua 67(3) (2021)

    Noreen, N., et al.: Brain tumor classification based on fine-tuned models and the ensemble method. Computers, Materials & Continua 67(3) (2021)

    2021

  30. [38]

    Computers in Biology and Medicine 191, 110214 (2025) 38

    Khan, S.U.R.: Multi-level feature fusion network for kidney disease detection. Computers in Biology and Medicine 191, 110214 (2025) 38

    2025

Pith tools

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