Pith. sign in

REVIEW 4 major objections 8 minor 1 cited by

Hierarchical Deep Feature Fusion and Ensemble Learning for Enhanced Brain Tumor MRI Classification

T0 review · 4 major / 8 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims that fusing deep features from two pre-trained Vision Transformers and classifying them with a tuned MLP achieves 99.83% accuracy on a binary brain-tumor MRI dataset, surpassing state-of-the-art methods.

desk verdict Reported 0.9983 accuracy is a selected maximum from many configurations on the same test split, so the SOTA claim isn't supported; the paper is a thorough but flawed empirical survey. read the letter →

arxiv 2506.12363 v1 pith:NOURJ55H submitted 2025-06-14 cs.CV

Zahid Ullah , Jihie Kim This is my paper
classification cs.CV
keywords braintumorclassificationMRIVisionTransformerfeatureensemblingclassifiertransferlearninghyperparameteroptimizationmedicalimageanalysis
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 sets out to show that a two-level ensemble—concatenating deep features from the best-performing pre-trained Vision Transformers and then feeding those combined features into tuned classical machine-learning classifiers—can classify brain MRI scans as tumorous or normal with near-perfect accuracy. On the larger of two public binary datasets, the paper reports that the combination of two ViT feature extractors with an MLP classifier reaches 0.9983 (99.83%) accuracy, and on the smaller dataset the best ViT pair with an MLP reaches 0.9589 (95.89%). The authors present this double ensembling as surpassing state-of-the-art methods and as evidence that feature fusion plus classifier fusion is a practical route to reliable computer-aided diagnosis without training a deep network from scratch. The numbers matter because they are claimed for frozen feature extractors plus relatively simple classifiers, which would be cheap to deploy in clinical settings.

What carries the argument

The carrying mechanism is a two-stage ensemble built on pre-trained ViT feature extractors, where a Vision Transformer (ViT) is a deep network that splits an image into patches and processes them with self-attention. Thirteen ViT variants first convert each MRI into a deep feature vector; nine ML classifiers—a multilayer perceptron (MLP), XGBoost, Gaussian Naive Bayes, AdaBoost, k-NN, Random Forest, and SVMs with linear, sigmoid, and RBF kernels—score those features. Models are ranked by average accuracy across the nine classifiers, with ties broken by lower standard deviation, and the top two or three feature sets are concatenated into a single vector. That fused vector goes into the classifiers again, and optionally the top classifiers are combined by majority voting, with preprocessing variants applied at different points. The concatenation is what the paper means by feature-level ensembling, and it is the step that produces the reported accuracy jumps.

What would settle it

Hold out the 20% test split before any model selection; choose the top two ViT feature extractors and the best classifier using only the training portion or a separate validation split, then evaluate once on the held-out set. If the chosen combination's accuracy on the larger dataset falls materially below the reported 0.9983, the central superiority claim is not supported.

Watch

Extended reading notes

Core claim

The central claim is that concatenating the features of two pre-trained ViT models—vit_large_patch16_224 and vit_base_patch32_384 on the larger dataset, and vit_base_patch16_224 and vit_small_patch32_224 on the smaller one—and classifying the fused vector with an MLP produces the best results, with a reported accuracy of 0.9983 (99.83%) on BT-large-2c and 0.9589 (95.89%) on BT-small-2c. The paper further claims that applying normalization, PCA, and SMOTE preprocessing, along with hyperparameter-optimized classifiers and majority-voting ensembles of classifiers, improves stability and often accuracy over single models. Its stated conclusion is that this hybrid DL-ML pipeline significantly surpasses state-of-the-art methods and that the combination of feature-level and classifier-level ensembling is the reason.

Load-bearing premise

The load-bearing premise is that choosing the best feature extractors, classifiers, and preprocessing variants from the 20% test set's accuracy tables does not inflate the final reported accuracies; if that selection has any bias, the headline numbers overstate real-world performance.

Editorial extensions

If this is right

  • If the reported numbers hold, a frozen ViT feature pair plus an MLP is enough for essentially perfect binary tumor/normal separation on these datasets, so no end-to-end fine-tuning is required.
  • Preprocessing and hyperparameter tuning are not optional extras: the paper's comparison tables show the average top accuracy on the larger dataset rising from about 0.9789 without preprocessing to about 0.9908 with it.
  • Classifier-level ensembles, especially MLP plus SVM variants, routinely match or beat their best single member, supporting the paper's claim that prediction fusion adds robustness.
  • The best feature extractors differ between datasets, so the paper's method includes a model-selection step that should be repeated per dataset rather than assumed transferable.

Reading between the lines

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

  • A reader should not take 99.83% as an unbiased estimate of future performance: because the same 20% test split was used to pick the winning feature/classifier combinations and then to report their accuracies, the true generalization gap is probably larger; a nested or held-out validation design would settle this.
  • The near-ceiling result on the larger dataset suggests that generic pre-trained features already separate these particular images, so the main contribution may be the selection protocol rather than a new learning mechanism; a simpler baseline should be checked under the same selection rules.
  • A direct extension would be to apply the same top-2 feature-concatenation recipe to multi-class tumor typing or to other imaging modalities, where the accuracy headroom is larger and the fusion benefit easier to detect.
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 / 8 minor

Summary. The paper proposes a double-ensembling pipeline for binary brain tumor classification from MRI: 13 frozen ImageNet-pretrained Vision Transformers extract features; nine scikit-learn classifiers (MLP, XGBoost, Gaussian NB, AdaBoost, k-NN, RF, SVM linear/sigmoid/RBF) are hyperparameter-tuned via grid search; top-2 or top-3 ViT feature sets are concatenated; and top classifiers are also combined by majority voting. Preprocessing variants (none, normalization+PCA, SMOTE, and all three combined) are compared. Experiments on two Kaggle datasets (BT-small-2c, 253 images, 202/51 train/test; BT-large-2c, 3000 images, 2400/600 train/test) report very high accuracies, including 0.9589 and 0.9983 for the best feature-level ensembles, and the abstract claims the approach significantly surpasses state-of-the-art methods.

Significance. If the reported methodology and numbers were established, the paper would offer a practical and reproducible recipe (frozen ViT features plus standard classifiers) and a notable empirical result: a two-ViT feature ensemble plus an MLP reaching near-perfect binary classification on these Kaggle benchmarks. Strengths include the explicit search spaces for all classifiers, the systematic ablation of preprocessing variants, and the promised public code repository. However, the central quantitative claim is contingent on an unbiased evaluation protocol, which the manuscript does not establish; without a held-out test set that is never used for model selection, the significance of the reported near-perfect accuracies and of the SOTA claim remains unsubstantiated.

major comments (4)
  1. [Sections 3.6, 4.1.2, and 5.3; Eq. (15)] The evaluation protocol has no untouched hold-out set, so the reported accuracies are likely inflated by selection on the test data. The dataset section (Section 3.2) reports only an 80/20 train/test split, yet Eq. (15) defines hyperparameter optimization as minimizing the loss on the test set X_Ts, and Section 3.6 states that the settings producing the best outcomes on the validation set are chosen without ever describing a validation split. Sections 4.1.2 and 5.3 select the top ViT feature extractors and the best feature/classifier/preprocessing combinations from accuracy tables computed on the same test set (Tables 5 through 15). The headline 0.9983 on BT-large-2c is therefore the maximum over a large grid (13 feature extractors, 9 classifiers, several preprocessing variants, and ensemble configurations) evaluated on the same 600-image test set. This selection process can inflate expected accuracy nontrivially on a 600-sample binary task; a final evaluation on a split not used for any model selection is required before the numbers can be taken as unbiased generalization estimates.
  2. [ Section 5.3 and Abstract] The central claim of surpassing state-of-the-art methods is not supported by any same-protocol baseline. Table 1 lists accuracies of prior works on datasets of different sizes, numbers of classes, and evaluation schemes, and the text never identifies which prior method is being outperformed on BT-large-2c or BT-small-2c under the same train/test split. The abstract's claim that the approach significantly surpasses SOTA is therefore unverified by the evidence presented. At minimum, the authors should run competing methods on the same preprocessed splits, or restrict the claim to relative gains over their own ablations.
  3. [Tables 8-15 and Section 5.3] The manuscript inconsistently labels proportions as percentages, which obscures the reported effect size and affects every headline number. For example, Section 5.3 reports 'a superior accuracy of 0.9983%' and the tables present values such as 0.9983 as if they were percentages, while the same values are proportions (0.9983 = 99.83%). Reporting 'accuracy of 0.9983%' is literally a claim below one percent. The text and tables must be corrected to a single unambiguous convention.
  4. [Sections 3.3 and 5.4] The preprocessing comparison is confounded with data augmentation. Section 3.3 applies rotation and horizontal flipping to the training data as part of the preprocessing pipeline, and Section 5.4 attributes the improvement from Table 5 to Table 6 to preprocessing (cropping, resizing, normalization, PCA). Since augmentation is present in the 'preprocessed' arm and absent in the 'non-preprocessed' arm, the observed gains cannot be attributed specifically to cropping or to the feature-level preprocessing variants (normalization/PCA/SMOTE). Please disaggregate the contributions of augmentation, cropping, normalization/PCA, and SMOTE, or temper the causal claims about preprocessing.
minor comments (8)
  1. [Section 3.2, Table 2] Table 2 gives only train and test counts; please also state the class balance of BT-small-2c after augmentation (the source dataset has 155 tumor and 98 non-tumor images) and specify whether the split was stratified.
  2. [Section 4.1.2] The tie-breaking rule refers to lower standard deviation across classifiers, but no standard deviations are reported in Tables 5-7, so the rule is not actually applied in a traceable manner.
  3. [Section 5.3, Tables 8-15] The tables would be much easier to check if the single highest value in each table were bolded and if each table were explicitly tied to the preprocessing variant described in the text; several paragraphs in Section 5.3 mix references to simple, normalization+PCA, and SMOTE results without a clean one-to-one mapping.
  4. [Section 3.5.1, Eqs. (1)-(2)] The notation is inconsistent: Eq. (1) defines y(l), Eq. (2) defines h(l) with an activation on z(l), but z(l) is never defined; please align the symbols.
  5. [Section 3.5.2, Eq. (9)] The formula for P(y|X) has a formatting error (a missing fraction bar) and should use class-conditional notation for the Gaussian parameters consistent with Eq. (11).
  6. [Section 3.5.5 and Table 4] The text states that Random Forest uses the square root of the number of features for splits, but Table 4 lists max_features as 'auto, sqrt, log2'; please reconcile the two descriptions.
  7. [Section 2.2] The description of reference [30] says the model is ResNet50 while the surrounding text and table describe it as a hybrid CNN approach; please check the description against the cited paper.
  8. [Abstract and Introduction] The binary nature of the task is not stated explicitly in the abstract; please say that both datasets are two-class, since this materially affects how the near-perfect accuracy numbers and SOTA comparisons are interpreted.

Circularity Check

2 steps flagged · score 6.0 of 10

The headline 0.9983% is not an independent prediction: Eq. (15) defines hyperparameter optimization as minimizing loss on the test set, and Tables 8–15 then select the best configuration from that same test set.

  1. fitted input called prediction [Section 3.6, Eq. (15); reported in Section 5.3 / Table 9]
    "The goal of HPO is to identify the optimal settings λast that produce an ideal model M ast, which minimizes the loss function L(XTs; M ). λ∗ = argmin λ L(XTs; A(XTr; λ)) = argmin λ F (XTs, XTr, A, λ, L) (15). Here, F represents the model’s objective function ... the datasets XTs(test set) and XTr(training set) are provided [55]."

    The paper describes only an 80/20 train/test split (Section 3.2), then explicitly defines the HPO objective as minimizing loss on that test set X_Ts. Every ML classifier reported as 'fine-tuned' is therefore selected to minimize the very test-set loss that is later reported as final accuracy. The headline 0.9983% is the minimized value of Eq. (15) for the chosen grid point, not an independent evaluation of a fixed model. Reporting the optimization criterion as the result makes the 'prediction' equivalent to the fitted objective by construction; without a separate validation split, no out-of-sample estimate remains.

  2. fitted input called prediction [Sections 4.1.2 and 5.3 (feature/ensemble selection from Tables 8–15)]
    "The evaluation process involved calculating the average accuracy achieved by each ViT model across nine different ML classifiers. ... The top three deep features are then utilized in our ensemble module. ... Table 9 ... the combination of 'vit large patch16 224 + vit base patch32 384' also paired with an MLP classifier, resulting in a superior accuracy of 0.9983%."

    Both the ranking of ViT extractors and the final reported 'best' combination are computed from the same 20% test split. The top-2/top-3 feature sets are chosen because their average accuracy across classifiers is highest on that split, and Table 9's 0.9983 is then reported as the result of the chosen combination. Selecting a configuration because its score is maximum in Tables 8–15 and then reporting that same score as the achieved accuracy is reporting a selection maximum, not testing a prediction; the reported value is forced by the selection rule.

full rationale

The central circularity is not self-citation: the only author self-citation, [26], is a related-work comparison of an older handcrafted-feature pipeline and is not load-bearing for the ViT ensemble results. The circularity is instead in the evaluation protocol. Eq. (15) explicitly sets the HPO objective to be the loss on the test set X_Ts, and Section 3.2 reports only an 80/20 train/test split with no separate validation split. The same test set is then used for feature-extractor selection (Section 4.1.2), hyperparameter tuning (Section 3.5), and final accuracy reporting (Section 5.3, Tables 8–15). The reported 0.9983% is therefore the maximum of a large grid of test-set scores after selecting on those scores, not an unbiased out-of-sample prediction. This is partial circularity—the ensemble idea itself is a legitimate empirical construction, but the headline quantitative claim reduces to the selection criterion by construction.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

No new physics or entities are introduced; the only inputs beyond data are hyperparameters tuned via grid search and the choice of ensemble members.

free parameters (5)
  • k-NN number of neighbors = 1 to 4 (best by accuracy)
    Selected by grid search on the validation set for each feature set (Section 3.5.4).
  • SVM C and gamma = C in [0.1, 1, 10, 100, 1000], gamma in [scale, auto, 0.1, 1, 10]
    Grid-searched per kernel (Section 3.5.6, Table 4).
  • RF n_estimators and max_depth = 100-500, None-50
    Grid-searched (Table 4).
  • XGBoost max_depth, learning_rate, subsample, n_estimators = [3,5,7], [0.1,0.01,0.001], [0.5,0.7,1], [100,200,300]
    Grid-searched (Table 4).
  • PCA components and SMOTE sampling ratio = not specified
    Used in preprocessing variants (Section 3.2 and 5.3), parameters not reported.
assumptions (3)
  • domain assumption The 80/20 random split yields i.i.d. training and test samples from the same distribution as clinical brain MRI.
    Invoked implicitly in Section 4.2 for the train/test split.
  • domain assumption ImageNet-pretrained ViT weights transfer to grayscale brain MRIs after resizing to 224/384 pixels.
    Used in Section 3.4 to justify frozen feature extraction.
  • domain assumption The Kaggle datasets contain correct binary labels.
    No manual audit of labels is described.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hierarchical Deep Feature Fusion and Ensemble Learning for Enhanced Brain Tumor MRI Classification." pith.science (2026). https://pith.science/paper/NOURJ55H

@misc{pith2026250612363,
  author       = {Pith},
  title        = {Pith review of: Hierarchical Deep Feature Fusion and Ensemble Learning for Enhanced Brain Tumor MRI Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NOURJ55H}},
  note         = {Machine review of arXiv:2506.12363}
}
read the original abstract

Accurate brain tumor classification is crucial in medical imaging to ensure reliable diagnosis and effective treatment planning. This study introduces a novel double ensembling framework that synergistically combines pre-trained deep learning (DL) models for feature extraction with optimized machine learning (ML) classifiers for robust classification. The framework incorporates comprehensive preprocessing and data augmentation of brain magnetic resonance images (MRI), followed by deep feature extraction using transfer learning with pre-trained Vision Transformer (ViT) networks. The novelty lies in the dual-level ensembling strategy: feature-level ensembling, which integrates deep features from the top-performing ViT models, and classifier-level ensembling, which aggregates predictions from hyperparameter-optimized ML classifiers. Experiments on two public Kaggle MRI brain tumor datasets demonstrate that this approach significantly surpasses state-of-the-art methods, underscoring the importance of feature and classifier fusion. The proposed methodology also highlights the critical roles of hyperparameter optimization (HPO) and advanced preprocessing techniques in improving diagnostic accuracy and reliability, advancing the integration of DL and ML for clinically relevant medical image analysis.

Figures

Figures reproduced from arXiv: 2506.12363 by the authors.

Figure 1
Figure 1. Workflow of our proposed double ensembling framework for brain tumor classi [PITH_FULL_IMAGE:figures/full_fig_p013_1.png] view at source ↗
Figure 2
Figure 2. The proposed double ensembling architecture for brain tumor classification, [PITH_FULL_IMAGE:figures/full_fig_p015_2.png] view at source ↗
Figure 3
Figure 3. MRIs with Corresponding Classes from Each Dataset. [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Pre-processing steps to crop the MRI images [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]
Figure 5
Figure 5. Figure 5: Methodology of transfer learning. with additional training samples. 3.4. Deep Feature Extraction using Pre-trained Visions Transformers We employed variants of ViT-based models as DL feature extractors, uti￾lizing their capacity to autonomously recognize and capture es…
Figure 6
Figure 6. Figure 6: Overview of ViT Architecture for Brain MRI analysis. [PITH_FULL_IMAGE:figures/full_fig_p018_6.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Systematic Integration of Attention Modules into CNNs for Accurate and Generalizable Medical Image Diagnosis

    cs.CV 2025-09 reject novelty 3.0 of 10

    Attention-augmented CNNs usually beat plain CNNs on two medical image datasets, with EfficientNetB5 plus hybrid attention the best, but test-set-based model selection undermines the claimed consistency.

Reference graph

Works this paper leans on

57 extracted references · 32 canonical work pages · cited by 1 Pith paper

  1. [1]

    B. Amin, R. S. Samir, Y. Tarek, M. Ahmed, R. Ibrahim, M. Ahmed, M. Hassan, Brain tumor multi classification and segmentation in mri images using deep learning, arXiv preprint arXiv:2304.10039 (2023)

    arXiv 2023

  2. [2]

    D. Xu, X. Wang, J. Cai, P.-A. Heng, Cross-modality guidance-aided multi-modal learning with dual attention for mri brain tumor grading, arXiv preprint arXiv:2401.09029 (2024)

    work page Pith review arXiv 2024

  3. [3]

    M. T. R, V. K. V, S. Guluwadi, Enhancing brain tumor detection in mri images through explainable ai using grad-cam with resnet 50, BMC medical imaging 24 (2024) 107

    work page 2024

  4. [4]

    Zulfiqar, U

    F. Zulfiqar, U. I. Bajwa, Y. Mehmood, Multi-class classification of brain tumor types from mr images using efficientnets, Biomedical Signal Processing and Control 84 (2023) 104777

    work page 2023

  5. [5]

    Ayg¨ un, Y

    M. Ayg¨ un, Y. H. S ¸ahin, G. ¨Unal, Multi modal convolutional neural networks for brain tumor segmentation, arXiv preprint arXiv:1809.06191 (2018)

    arXiv 2018

  6. [6]

    A. A. Dehkordi, M. Hashemi, M. Neshat, S. Mirjalili, A. S. Sadiq, Brain tumor detection and classification using a new evolutionary convolu- tional neural network, arXiv preprint arXiv:2204.12297 (2022). 41

    work page Pith review arXiv 2022

  7. [7]

    H. P. A. Tjahyaningtijas, Brain tumor image segmentation in mri image, in: IOP Conference series: materials science and engineering, volume 336, IOP Publishing, 2018, p. 012012

    work page 2018

  8. [8]

    Madgi, S

    M. Madgi, S. Giraddi, G. Bharamagoudar, M. Madhur, Brain tumor classification and segmentation using deep learning, in: Smart Com- puting Techniques and Applications: Proceedings of the Fourth Inter- national Conference on Smart Computing and Informatics, Volume 2, Springer, 2021, pp. 201–208

    work page 2021

Show all 57 references
  1. [9]

    S. R. Gunasekara, H. Kaldera, M. B. Dissanayake, A feasibility study for deep learning based automated brain tumor segmentation using mag- netic resonance images, arXiv preprint arXiv:2012.11952 (2020)

    2020 arXiv

  2. [10]

    Gundogan, A novel hybrid deep learning model enhanced with ex- plainable ai for brain tumor multi-classification from mri images, Ap- plied Sciences 15 (2025) 5412

    E. Gundogan, A novel hybrid deep learning model enhanced with ex- plainable ai for brain tumor multi-classification from mri images, Ap- plied Sciences 15 (2025) 5412

    2025

  3. [11]

    M. K. Abd-Ellah, A. I. Awad, A. A. Khalaf, H. F. Hamed, A re- view on brain tumor diagnosis from mri images: Practical implications, key achievements, and lessons learned, Magnetic resonance imaging 61 (2019) 300–318

    2019

  4. [12]

    Magadza, S

    T. Magadza, S. Viriri, Deep learning for brain tumor segmentation: a survey of state-of-the-art, Journal of Imaging 7 (2021) 19

    2021

  5. [13]

    M. W. Nadeem, M. A. A. Ghamdi, M. Hussain, M. A. Khan, K. M. Khan, S. H. Almotiri, S. A. Butt, Brain tumor analysis empowered with deep learning: A review, taxonomy, and future challenges, Brain sciences 10 (2020) 118

    2020

  6. [14]

    Faradibah, D

    A. Faradibah, D. Widyawati, A. U. T. Syahar, S. R. Jabir, P. L. L. Bel- luano, Comparison analysis of random forest classifier, support vector machine, and artificial neural network performance in multiclass brain tumor classification, Indonesian Journal of Data and Science 4 ...

    2023

  7. [15]

    Latif, G

    G. Latif, G. Ben Brahim, D. A. Iskandar, A. Bashar, J. Alghazo, Glioma tumors’ classification using deep-neural-network-based features with svm classifier, Diagnostics 12 (2022) 1018. 42

    2022

  8. [16]

    Ahmad, P

    S. Ahmad, P. K. Choudhury, On the performance of deep transfer learn- ing networks for brain tumor detection using mr images, IEEE Access 10 (2022) 59099–59114

    2022

  9. [17]

    Takahashi, Y

    S. Takahashi, Y. Sakaguchi, N. Kouno, K. Takasawa, K. Ishizu, Y. Akagi, R. Aoyama, N. Teraya, A. Bolatkan, N. Shinkai, et al., Comparison of vision transformers and convolutional neural networks in medical image analysis: a systematic review, Journal of Medical Systems 48 (2024) 84

    2024

  10. [18]

    Matsoukas, J

    C. Matsoukas, J. F. Haslum, M. S¨ oderberg, K. Smith, Pretrained vits yield versatile representations for medical images, arXiv preprint arXiv:2303.07034 (2023)

    2023 arXiv

  11. [19]

    Dahan, A

    S. Dahan, A. Fawaz, L. Z. Williams, C. Yang, T. S. Coalson, M. F. Glasser, A. D. Edwards, D. Rueckert, E. C. Robinson, Surface vision transformers: Attention-based modelling applied to cortical analysis, in: International Conference on Medical Imaging with Deep Learning, PMLR,...

    2022

  12. [20]

    C.-M. Feng, Y. Yan, G. Chen, Y. Xu, Y. Hu, L. Shao, H. Fu, Multimodal transformer for accelerated mr imaging, IEEE Transactions on Medical Imaging 42 (2022) 2804–2816

    2022

  13. [21]

    Dahan, L

    S. Dahan, L. Z. Williams, D. Rueckert, E. C. Robinson, The multiscale surface vision transformer, ArXiv (2024) arXiv–2303

    2024

  14. [22]

    G. K. Thakur, A. Thakur, S. Kulkarni, N. Khan, S. Khan, Deep learning approaches for medical image analysis and diagnosis, Cureus 16 (2024)

    2024

  15. [23]

    Babayomi, O

    M. Babayomi, O. A. Olagbaju, A. A. Kadiri, Convolutional xg- boost (c-xgboost) model for brain tumor detection, arXiv preprint arXiv:2301.02317 (2023)

    2023 arXiv

  16. [24]

    G. Zhu, B. Jiang, L. Tong, Y. Xie, G. Zaharchuk, M. Wintermark, Applications of deep learning to neuro-imaging techniques, Frontiers in neurology 10 (2019) 869

    2019

  17. [25]

    B. Ural, A computer-based brain tumor detection approach with ad- vanced image processing and probabilistic neural network methods, Journal of Medical and Biological Engineering 38 (2018) 867–879. 43

    2018

  18. [26]

    Ullah, M

    Z. Ullah, M. U. Farooq, S.-H. Lee, D. An, A hybrid image enhancement based brain mri images classification technique, Medical hypotheses 143 (2020) 109922

    2020

  19. [27]

    Varuna Shree, T

    N. Varuna Shree, T. Kumar, Identification and classification of brain tumor mri images with feature extraction using dwt and probabilistic neural network, Brain informatics 5 (2018) 23–30

    2018

  20. [28]

    Kharrat, K

    A. Kharrat, K. Gasmi, M. B. Messaoud, N. Benamrane, M. Abid, A hybrid approach for automatic classification of brain mri using genetic algorithm and support vector machine, Leonardo journal of sciences 17 (2010) 71–82

    2010

  21. [29]

    Rajan, C

    P. Rajan, C. Sundar, Brain tumor detection and segmentation by in- tensity adjustment, Journal of medical systems 43 (2019) 282

    2019

  22. [30]

    C ¸ inar, M

    A. C ¸ inar, M. Yildirim, Detection of tumors on brain mri images using the hybrid convolutional neural network architecture, Medical hypothe- ses 139 (2020) 109684

    2020

  23. [31]

    Mehnatkesh, S

    H. Mehnatkesh, S. M. J. Jalali, A. Khosravi, S. Nahavandi, An intel- ligent driven deep residual learning framework for brain tumor classifi- cation using mri images, Expert Systems with Applications 213 (2023) 119087

    2023

  24. [32]

    Deepak, P

    S. Deepak, P. Ameer, Brain tumor classification using deep cnn features via transfer learning, Computers in biology and medicine 111 (2019) 103345

    2019

  25. [33]

    F. J. D ´ ıaz-Pernas, M. Mart ´ ınez-Zarzuela, M. Ant´ on-Rodr ´ ıguez, D. Gonz´ alez-Ortega, A deep learning approach for brain tumor clas- sification and segmentation using a multiscale convolutional neural net- work, in: Healthcare, volume 9, MDPI, 2021, p. 153

    2021

  26. [34]

    M. S. I. Khan, A. Rahman, T. Debnath, M. R. Karim, M. K. Nasir, S. S. Band, A. Mosavi, I. Dehzangi, Accurate brain tumor detection using deep convolutional neural network, Computational and structural biotechnology journal 20 (2022) 4733–4745

    2022

  27. [35]

    J. S. Paul, A. J. Plassard, B. A. Landman, D. Fabbri, Deep learning for brain tumor classification, in: Medical Imaging 2017: Biomedical 44 Applications in Molecular, Structural, and Functional Imaging, volume 10137, SPIE, 2017, pp. 253–268

    2017

  28. [36]

    D. J. Hemanth, J. Anitha, A. Naaji, O. Geman, D. E. Popescu, L. H. Son, A modified deep convolutional neural network for abnormal brain image classification, IEEE Access 7 (2018) 4275–4283

    2018

  29. [37]

    Y. Shen, P. Guo, J. Wu, Q. Huang, N. Le, J. Zhou, S. Jiang, M. Un- berath, Movit: Memorizing vision transformers for medical image anal- ysis, in: International Workshop on Machine Learning in Medical Imag- ing, Springer, 2023, pp. 205–213

    2023

  30. [38]

    K. Xia, J. Wang, Recent advances of transformers in medical image analysis: a comprehensive review, MedComm–Future Medicine 2 (2023) e38

    2023

  31. [39]

    E. U. Henry, O. Emebob, C. A. Omonhinmin, Vision transformers in medical imaging: A review, arXiv preprint arXiv:2211.10043 (2022)

    2022 arXiv

  32. [40]

    Chakrabarty, Brain mri images for brain tumor detection, J

    N. Chakrabarty, Brain mri images for brain tumor detection, J. Exp. Med 216 (2019) 539–555

    2019

  33. [41]

    Hamada, Br35h brain tumor detection 2020 dataset, https://www.kaggle.com/ahmedhamada0/braintumor-detection (2020)

    A. Hamada, Br35h brain tumor detection 2020 dataset, https://www.kaggle.com/ahmedhamada0/braintumor-detection (2020)

    2020

  34. [42]

    Zhang, X

    X. Zhang, X. Zhou, M. Lin, J. Sun, Finding extreme points in contours with opencv, PyImageSearch. Available online: https://www. pyimagesearch. com/2016/04/11/finding-extreme-points- in-contours-with-opencv (accessed on 10 August 2020) (2020)

    2020

  35. [43]

    Perez, J

    L. Perez, J. Wang, The effectiveness of data augmentation in image clas- sification using deep learning, arXiv preprint arXiv:1712.04621 (2017)

    2017 arXiv

  36. [44]

    S. Yang, W. Xiao, M. Zhang, S. Guo, J. Zhao, F. Shen, Image data aug- mentation for deep learning: A survey, arXiv preprint arXiv:2204.08610 (2022)

    2022 arXiv

  37. [45]

    Dosovitskiy, L

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, et al., An image is worth 16x16 words: Transformers for image recognition at scale, arXiv preprint arXiv:2010.11929 (2020). 45

    2020 arXiv

  38. [46]

    B. Wu, C. Xu, X. Dai, A. Wan, P. Zhang, Z. Yan, M. Tomizuka, J. Gon- zalez, K. Keutzer, P. Vajda, Visual transformers: Token-based im- age representation and processing for computer vision, arXiv preprint arXiv:2006.03677 (2020)

    2020 arXiv

  39. [47]

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, L. Fei-Fei, Imagenet: A large-scale hierarchical image database, in: 2009 IEEE conference on computer vision and pattern recognition, Ieee, 2009, pp. 248–255

    2009

  40. [48]

    B. J. Erickson, P. Korfiatis, Z. Akkus, T. L. Kline, Machine learning for medical imaging, radiographics 37 (2017) 505–515

    2017

  41. [49]

    X. Meng, J. Jiang, H. Wang, Agwo: Advanced gwo in multi-layer per- ception optimization, Expert Systems with Applications 173 (2021) 114676

    2021

  42. [50]

    Freund, R

    Y. Freund, R. E. Schapire, A decision-theoretic generalization of on- line learning and an application to boosting, Journal of computer and system sciences 55 (1997) 119–139

    1997

  43. [51]

    Breiman, Random forests, Machine learning 45 (2001) 5–32

    L. Breiman, Random forests, Machine learning 45 (2001) 5–32

    2001

  44. [52]

    Cortes, V

    C. Cortes, V. Vapnik, Support-vector networks, Machine learning 20 (1995) 273–297

    1995

  45. [53]

    T. Yu, H. Zhu, Hyper-parameter optimization: A review of algorithms and applications, arXiv preprint arXiv:2003.05689 (2020)

    2020 arXiv

  46. [54]

    Tran, J.-G

    N. Tran, J.-G. Schneider, I. Weber, A. K. Qin, Hyper-parameter opti- mization in classification: To-do or not-to-do, Pattern Recognition 103 (2020) 107245

    2020

  47. [55]

    Claesen, B

    M. Claesen, B. De Moor, Hyperparameter search in machine learning, arXiv preprint arXiv:1502.02127 (2015)

    2015 arXiv

  48. [56]

    D. M. Belete, M. D. Huchaiah, Grid search in hyperparameter optimiza- tion of machine learning models for prediction of hiv/aids test results, International Journal of Computers and Applications 44 (2022) 875–886

    2022

  49. [57]

    Krizhevsky, I

    A. Krizhevsky, I. Sutskever, G. E. Hinton, Imagenet classification with deep convolutional neural networks, Advances in neural information processing systems 25 (2012). 46

    2012

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.