Pith. sign in

REVIEW 4 major objections 5 minor 27 references

A Comprehensive Analysis of COVID-19 Detection Using Bangladeshi Data and Explainable AI

T0 review · 4 major / 5 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 A VGG19 model trained on SMOTE-oversampled data reaches 98% accuracy on a four-class Bangladeshi chest X-ray set, with LIME explaining the predictions.

desk verdict Routine but potentially careless benchmark: the 98% VGG19 accuracy on Bangladeshi CXR data hinges on an ambiguous SMOTE placement that could leak test information into training, and a single split with no error bars can't support fine-grained model comparisons. read the letter →

arxiv 2506.07234 v1 pith:UC7RRJBV submitted 2025-06-08 eess.IV cs.CV

Shuvashis Sarker This is my paper
classification eess.IVcs.CV
keywords COVID-19detectionchestX-raytransferlearningVGG19SMOTEexplainableAILIMEBangladeshidataset
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 reports that a VGG19 transfer learning model, trained after SMOTE oversampling to 1,500 images per class, reaches 98% accuracy on a four-class chest X-ray dataset collected in Bangladesh (Normal, Lung Opacity, COVID-19, and Viral Pneumonia). It argues that this beats plain deep learning models (CNN, BiLSTM) and classical machine learning (SVM, Random Forest), and that LIME explanations reveal the image regions driving both correct and incorrect predictions. The study's goal is to show that Bangladeshi CXR data can support accurate, transparent automated COVID-19 screening that works in resource-limited settings.

What carries the argument

The central objects are SMOTE (Synthetic Minority Oversampling Technique), which generates synthetic feature-space instances to balance the classes; VGG19, a pre-trained CNN with added dense layers and a BiLSTM layer for sequential learning; a multistage preprocessing chain (grayscale conversion, Laplacian filtering, sharpening, Sobel gradients, and gamma correction); and LIME, which builds local interpretable approximations to explain individual predictions. The load-bearing piece is the SMOTE-oversampled training set at 1,500 samples per class: the paper attributes the 98% accuracy to this balance, and the comparison of SMOTE-1 versus SMOTE-2 is the evidence that the oversampling level matters.

What would settle it

Train the identical VGG19 pipeline but apply SMOTE after the train/validation/test split, generating synthetic samples only from the training folds; if accuracy falls materially below 98%, the headline number was inflated by leakage. A second check is to inspect the k-nearest-neighbor indices used during SMOTE synthesis to confirm that no synthetic instance is derived from a test-set image or shares a near-duplicate with one.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central discovery is that a transfer learning pipeline—VGG19 with added dense layers, fed preprocessed CXR images and balanced with SMOTE to 1,500 synthetic samples per class—classifies the four disease categories with 98% accuracy, 99% precision and recall, and a 0.98 F1 score. The paper also claims that SMOTE oversampling improves all model families tested, with the 1,500-per-class balance (SMOTE-2) outperforming the 1,200-per-class balance (SMOTE-1), and that LIME's superpixel explanations can indicate why a misclassification occurred, making the model's behavior inspectable.

Load-bearing premise

The 98% accuracy is only a valid measure of generalization if SMOTE oversampling is restricted to the training portion of the data; the paper's description of applying synthetic instances before the dataset split suggests this condition may not hold.

Editorial extensions

If this is right

  • If the 98% accuracy is genuine, the same pipeline could be retrained on larger Bangladeshi CXR collections to build a low-cost triage tool for hospitals without PCR capacity.
  • The reported trend (transfer learning outperforming deep learning, which outperforms classical ML) suggests VGG-style transfer learning should be the default baseline for similar four-class respiratory X-ray tasks.
  • The improvement from SMOTE-1 to SMOTE-2 implies there is a useful range of oversampling ratios before overfitting sets in, so future studies should tune rather than fix this parameter.
  • LIME's demonstrated ability to show why individual images are misclassified could be used in practice to flag uncertain or contradictory model outputs for radiologist review.
  • The multi-stage preprocessing chain may account for part of the gain, meaning future work should ablate each filter to determine which step contributes most.

Reading between the lines

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

  • The paper reports a single accuracy point estimate on one train/test split without cross-validation or confidence intervals, so the 98% figure should be read as a preliminary result until replicated across multiple splits.
  • Because the dataset has only 4,350 images from a single national source, the claimed accuracy may not transfer to other Bangladeshi hospitals, imaging devices, or patient demographics.
  • The description in Section 4.2 that synthetic instances are 'applied to the dataset' before the split raises a direct test: if SMOTE runs before splitting, test-derived synthetic images enter training, and the 98% accuracy would overstate real generalization; a post-split SMOTE re-run is the cleanest way to check.
  • A natural extension the author does not pursue is to report per-class recall and confusion matrices, which would show whether the high accuracy hides weaker performance on specific classes such as Viral Pneumonia.
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 / 5 minor

Summary. The paper reports a comparative study of machine learning, deep learning, and transfer learning models for classifying COVID-19 from a Bangladeshi chest X-ray dataset of 4,350 images across four classes (Normal, Lung-Opacity, COVID-19, Viral-Pneumonia). The authors apply an image preprocessing pipeline (grayscale, Laplacian filtering, sharpening, Sobel edge combination, power-law transformation), use SMOTE to oversample classes to 1,200 (SMOTE-1) and 1,500 (SMOTE-2) samples per class, and evaluate SVM, Random Forest, CNN, BiLSTM, VGG-16, and VGG-19. They report that VGG-19 achieves 98% accuracy on the SMOTE-2 dataset, and they use LIME to provide visual explanations for correct and incorrect classifications. The abstract and conclusion emphasize the 98% VGG-19 result as the main contribution.

Significance. If the headline result is methodologically sound, the paper would offer a useful benchmark for COVID-19 detection on a Bangladeshi chest X-ray dataset and a practical demonstration of LIME-based interpretability in a low-resource setting. The paper's strengths include the use of a local dataset, a broad comparison across model families, and the inclusion of LIME explanations for both correct and incorrect predictions. However, the significance is substantially limited by the lack of reproducible experimental details and by ambiguities in the data handling pipeline that directly affect the validity of the 98% accuracy claim. The paper does not provide code, split indices, seed values, or hyperparameter settings, and the SMOTE description is consistent with a leakage-prone procedure. As it stands, the central claim is not independently verifiable.

major comments (4)
  1. [Section 4.2, SMOTE] The SMOTE procedure is described as two-phase: 'first, I oversampled the minority class to achieve balance; second, I applied the generated synthetic instances to the dataset' (Section 4.2). This wording implies that SMOTE is applied to the entire dataset before the 80/10/10 split described in Section 4.1, rather than to the training partition only. If synthetic instances are generated from feature-space neighbors that include test images, and if those synthetic instances are then added to both the training and test sets, the reported 98% accuracy for VGG-19 (Table 1) is inflated and does not measure performance on real patient images. The manuscript must explicitly state the order of operations (split first, then SMOTE on training data only) and, if that is the intended pipeline, the current text must be corrected to remove the ambiguous phrase 'applied the generated synthetic instances to the dataset.' This issue is load-bearing because the entire headline claim depends on a clean separation between training and test data.
  2. [Section 5, Table 1] All results are reported from a single train/test split with no error bars, confidence intervals, or repeated runs. With a test set of approximately 10% of the data (roughly 435 images before SMOTE augmentation), the 2% difference between VGG-19 (98%) and VGG-16 (96%) may be within sampling variability, especially after SMOTE changes the test set composition. The conclusion that VGG-19 is the best model is therefore not statistically supported. The authors should provide results averaged over multiple splits (e.g., 5-fold cross-validation) or at least report the standard deviation across three to five independent runs, along with the exact number of test samples used per condition.
  3. [Section 4.1, Data Preprocessing] The preprocessing pipeline is underspecified. Equation (vii) defines the power-law transformation as I' = 255 * (I/255)^gamma, but the value of gamma used in the experiments is never given. Similarly, the kernel sizes for the Laplacian and Sobel filters are not stated, and it is unclear whether the 'sharpen' kernel is a standard 3x3 kernel. These parameters determine the input representation to all models and directly affect the reported accuracy; without them, the experiments cannot be replicated or meaningfully compared with other studies. The authors should provide the exact preprocessing configuration, ideally as a table or pseudocode.
  4. [Section 4.2 and Figure 2] The manuscript does not specify the feature space in which SMOTE operates. SMOTE is a feature-space method, but the paper applies it to images; it is unclear whether the synthetic examples are generated from raw pixel vectors, HOG features, or some other representation. If SMOTE is applied to raw pixels, the synthetic 'images' are unlikely to be realistic and may not resemble chest X-rays, yet they are then fed to VGG-19, whose pretrained weights expect natural image statistics. This point must be clarified, as it affects both the validity of the oversampling and the interpretability of the LIME results.
minor comments (5)
  1. [Section 1, Abstract and Introduction] The abstract contains 'Deep Learning(ML)' where it should read 'Deep Learning(DL)'. There are also grammatical errors such as 'gained significant interest because to its superior performance' in the Introduction; these should be corrected.
  2. [Section 4.2, SMOTE] The terms 'SMOTE-1' and 'SMOTE-2' are not standard; they appear to denote target sample counts of 1,200 and 1,500 per class. The authors should make this explicit in the text so readers do not mistake them for different SMOTE variants.
  3. [Section 5, Result Analysis] The sentence 'excessive oversampling can cause overfitting' is presented without evidence. Since the results show that SMOTE-2 (1,500 samples) achieves higher accuracy than SMOTE-1 (1,200 samples), the opposite trend is observed in this study. The authors should either substantiate this claim with experimental evidence (e.g., training curves or validation loss) or remove it.
  4. [Figure 2, Proposed Model Architecture] The figure is difficult to interpret: the label '??' appears, and the placement of SMOTE relative to the dataset split is not visually clear. The authors should redraw the figure to show the exact data flow, especially whether SMOTE occurs before or after the split.
  5. [Section 7, Conclusion & Future Work] The conclusion contains a typo, 'In conlusion,' and the final sentence about extending the dataset is vague. More importantly, the conclusion states 'The VGG19 model attains the utmost performance' without acknowledging the statistical concerns raised above; a more cautious phrasing is warranted.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the VGG19 98% claim is an empirical measurement, not a derivation from its own inputs.

full rationale

The paper's central claim is an accuracy measurement on a four-class Bangladeshi chest X-ray dataset after SMOTE oversampling. This is an empirical evaluation, not a derivation chain in which an output quantity is defined in terms of the claimed result. The VGG19 98% figure is reported from Table 1 as a measured value, and the comparison across Original, SMOTE-1 (1,200 samples/class), and SMOTE-2 (1,500 samples/class) is an explicit model-selection comparison rather than a hidden circular reduction. The paper does not invoke a self-citation as load-bearing evidence, does not import a uniqueness theorem from its own prior work, and does not rename a known result under new coordinates. The only notable concern is methodological: Section 4.2 describes SMOTE as applied to 'the dataset' before an explicit train/test split, which could imply that synthetic instances derived from test images enter training, inflating the reported accuracy. However, that is a data-handling validity risk and potential leakage, not circularity in the sense of a claim being equivalent to its input by construction. Because the accuracy is presented as a measured outcome and the SMOTE target sizes are not fitted parameters that define the reported metric, the circularity score is 0.

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

The central claim depends on the correctness of the dataset labels, on the assumption that SMOTE is applied without leaking test information, and on several hand-chosen preprocessing and oversampling parameters. No invented entities are introduced.

free parameters (3)
  • SMOTE-2 oversampling target per class = 1500
    The target of 1,500 samples per class is chosen after comparing 1,200 and 1,500; the higher value yields the best accuracy and is presented as the headline result.
  • Gamma exponent in power-law transformation = not specified
    The preprocessing uses a power-law transform with gamma, but the value is not reported; it is a hand-chosen parameter affecting input images.
  • k_neighbors in SMOTE = default
    The paper leaves SMOTE's k_neighbors at default, which is a modeling choice that influences synthetic sample quality.
assumptions (3)
  • domain assumption The Mendeley Bangladeshi CXR dataset labels are correct and the images are representative of the target population.
    The entire evaluation assumes ground-truth labels are accurate; no clinician verification is mentioned.
  • domain assumption Applying SMOTE before the train/test split is valid, or equivalently that no synthetic samples derived from test data enter training.
    The paper does not specify that SMOTE is restricted to the training split; if it is applied to the full dataset, the accuracy estimate is biased.
  • domain assumption The hand-crafted preprocessing (Laplacian, Sobel, gamma) preserves diagnostic information in CXR images.
    The preprocessing is applied to all images but its effect on model performance is not ablated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Comprehensive Analysis of COVID-19 Detection Using Bangladeshi Data and Explainable AI." pith.science (2026). https://pith.science/paper/UC7RRJBV

@misc{pith2026250607234,
  author       = {Pith},
  title        = {Pith review of: A Comprehensive Analysis of COVID-19 Detection Using Bangladeshi Data and Explainable AI},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UC7RRJBV}},
  note         = {Machine review of arXiv:2506.07234}
}
read the original abstract

COVID-19 is a rapidly spreading and highly infectious virus which has triggered a global pandemic, profoundly affecting millions across the world. The pandemic has introduced unprecedented challenges in public health, economic stability, and societal structures, necessitating the implementation of extensive and multifaceted health interventions globally. It had a tremendous impact on Bangladesh by April 2024, with around 29,495 fatalities and more than 2 million confirmed cases. This study focuses on improving COVID-19 detection in CXR images by utilizing a dataset of 4,350 images from Bangladesh categorized into four classes: Normal, Lung-Opacity, COVID-19 and Viral-Pneumonia. ML, DL and TL models are employed with the VGG19 model achieving an impressive 98% accuracy. LIME is used to explain model predictions, highlighting the regions and features influencing classification decisions. SMOTE is applied to address class imbalances. By providing insight into both correct and incorrect classifications, the study emphasizes the importance of XAI in enhancing the transparency and reliability of models, ultimately improving the effectiveness of detection from CXR images.

Figures

Figures reproduced from arXiv: 2506.07234 by the authors.

Figure 1
Figure 1. Comparison of CXR Images Before and After Preprocessing [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Proposed Model Architecture 5 Result Analysis Table1 presents a detailed comparison of model performance across three dataset condition: the Original Dataset, Smote 1, and Smote 2. In the Original one, the highest result is achieved by VGG 16 and VGG 19 models, both scoring 92%, demonstrating their superior performance among the models tested. In the Smote 1 dataset, the VGG 16 model has acheived with an accuracy of… view at source ↗
Figure 3
Figure 3. Some Sample Images of Actual Class Predicted Incorrectly by TL Model [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Some Sample Images of Actual Class Predicted correctly by TL Model [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

27 extracted references · 25 canonical work pages

  1. [1]

    Pneumonia classification using deep learning from chest x-ray images during covid-19.Cognitive Computation, pages 1–13, 2021

    Abdullahi Umar Ibrahim, Mehmet Ozsoz, Sertan Serte, Fadi Al-Turjman, and Polycarp Shizawaliyi Yakoi. Pneumonia classification using deep learning from chest x-ray images during covid-19.Cognitive Computation, pages 1–13, 2021

    work page 2021

  2. [2]

    Deep learning algorithm for covid-19 classification using chest x-ray images.Computational and Mathematical Methods in Medicine, 2021(1):9269173, 2021

    Sharmila VJ. Deep learning algorithm for covid-19 classification using chest x-ray images.Computational and Mathematical Methods in Medicine, 2021(1):9269173, 2021

    work page 2021

  3. [3]

    Analysis of covid-19 data using advanced machine learning models

    Callan Noak and Sujing Wang. Analysis of covid-19 data using advanced machine learning models. In2022 3rd International Conference on Pattern Recognition and Machine Learning (PRML), pages 430–437. IEEE, 2022

    work page 2022

  4. [4]

    A machine learning-based framework for diagnosis of covid-19 from chest x-ray images.Interdisciplinary Sciences: Computational Life Sciences, 13:103–117, 2021

    Jawad Rasheed, Alaa Ali Hameed, Chawki Djeddi, Akhtar Jamil, and Fadi Al-Turjman. A machine learning-based framework for diagnosis of covid-19 from chest x-ray images.Interdisciplinary Sciences: Computational Life Sciences, 13:103–117, 2021

    work page 2021

  5. [5]

    Covid detection from cxr scans using deep multi-layered cnn

    Rajarshi Bhadra and Subhajit Kar. Covid detection from cxr scans using deep multi-layered cnn. In2020 IEEE Bombay section signature conference (IBSSC), pages 214–218. IEEE, 2020

    work page 2020

  6. [6]

    Empowering covid-19 detection: Optimizing performance through fine-tuned efficientnet deep learning architecture.Computers in Biology and Medicine, 168:107789, 2024

    Md Alamin Talukder, Md Abu Layek, Mohsin Kazi, Md Ashraf Uddin, and Sunil Aryal. Empowering covid-19 detection: Optimizing performance through fine-tuned efficientnet deep learning architecture.Computers in Biology and Medicine, 168:107789, 2024

    work page 2024

  7. [7]

    Pre-trained xception model-based covid detection using cxr images

    M Gayathri Lakshmi, Komal Singh, R Beaulah Jeyavathana, V Satish Goud, A Deepak, et al. Pre-trained xception model-based covid detection using cxr images. In2022 International Conference on Automation, Computing and Renewable Systems (ICACRS), pages 1059–1063. IEEE, 2022

    work page 2022

  8. [8]

    Applications of machine learning algorithms to support covid-19 diagnosis using x-rays data information

    Elias P Medeiros, Marcos R Machado, Emannuel Diego G de Freitas, Daniel S da Silva, and Renato William R de Souza. Applications of machine learning algorithms to support covid-19 diagnosis using x-rays data information. Expert Systems with Applications, 238:122029, 2024

    work page 2024

Show all 27 references
  1. [9]

    Covid-19 ct-images diagnosis and severity assessment using machine learning algorithm.Cluster Computing, 27(1):547–562, 2024

    Zaid Albataineh, Fatima Aldrweesh, and Mohammad A Alzubaidi. Covid-19 ct-images diagnosis and severity assessment using machine learning algorithm.Cluster Computing, 27(1):547–562, 2024

    2024

  2. [10]

    Tawsifur Rahman, Muhammad EH Chowdhury, Amith Khandakar, Zaid Bin Mahbub, Md Sakib Abrar Hossain, Abraham Alhatou, Eynas Abdalla, Sreekumar Muthiyal, Khandaker Farzana Islam, Saad Bin Abul Kashem, et al. Bio-cxrnet: A robust multimodal stacking machine learning technique for m...

    2023

  3. [11]

    Efficient gan-based chest radiographs (cxr) augmentation to diagnose coronavirus disease pneumo- nia.International journal of medical sciences, 17(10):1439, 2020

    Saleh Albahli. Efficient gan-based chest radiographs (cxr) augmentation to diagnose coronavirus disease pneumo- nia.International journal of medical sciences, 17(10):1439, 2020

    2020

  4. [12]

    Unveil- ing covid-19 from chest x-ray with deep learning: a hurdles race with small data.International Journal of Environmental Research and Public Health, 17(18):6933, 2020

    Enzo Tartaglione, Carlo Alberto Barbano, Claudio Berzovini, Marco Calandri, and Marco Grangetto. Unveil- ing covid-19 from chest x-ray with deep learning: a hurdles race with small data.International Journal of Environmental Research and Public Health, 17(18):6933, 2020. 10 AP...

    2020

  5. [13]

    Rezaul Karim, Till Döhmen, Michael Cochez, Oya Beyan, Dietrich Rebholz-Schuhmann, and Stefan Decker

    Md. Rezaul Karim, Till Döhmen, Michael Cochez, Oya Beyan, Dietrich Rebholz-Schuhmann, and Stefan Decker. Deepcovidexplainer: Explainable covid-19 diagnosis from chest x-ray images. In2020 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pages 1034–1037, 2020

    2020

  6. [14]

    Covid-net cxr-2: An enhanced deep convolutional neural network design for detection of covid-19 cases from chest x-ray images.Frontiers in Medicine, 9:861680, 2022

    Maya Pavlova, Naomi Terhljan, Audrey G Chung, Andy Zhao, Siddharth Surana, Hossein Aboutalebi, Hayden Gunraj, Ali Sabri, Amer Alaref, and Alexander Wong. Covid-net cxr-2: An enhanced deep convolutional neural network design for detection of covid-19 cases from chest x-ray imag...

    2022

  7. [15]

    Weakly labeled data augmentation for deep learning: a study on covid-19 detection in chest x-rays.Diagnostics, 10(6):358, 2020

    Sivaramakrishnan Rajaraman and Sameer Antani. Weakly labeled data augmentation for deep learning: a study on covid-19 detection in chest x-rays.Diagnostics, 10(6):358, 2020

    2020

  8. [16]

    Alderson, Lucas S

    Sivaramakrishnan Rajaraman, Jenifer Siegelman, Philip O. Alderson, Lucas S. Folio, Les R. Folio, and Sameer K. Antani. Iteratively pruned deep learning ensembles for covid-19 detection in chest x-rays.IEEE Access, 8:115041– 115050, 2020

    2020

  9. [17]

    Deep gru-cnn model for covid-19 detection from chest x-rays data.IEEE Access, 10:35094–35105, 2022

    Pir Masoom Shah, Faizan Ullah, Dilawar Shah, Abdullah Gani, Carsten Maple, Yulin Wang, Shahid, Mohammad Abrar, and Saif Ul Islam. Deep gru-cnn model for covid-19 detection from chest x-rays data.IEEE Access, 10:35094–35105, 2022

    2022

  10. [18]

    Svd-clahe boosting and balanced loss function for covid-19 detection from an imbalanced chest x-ray dataset.Computers in Biology and Medicine, 150:106092, 2022

    Santanu Roy, Mrinal Tyagi, Vibhuti Bansal, and Vikas Jain. Svd-clahe boosting and balanced loss function for covid-19 detection from an imbalanced chest x-ray dataset.Computers in Biology and Medicine, 150:106092, 2022

    2022

  11. [19]

    Covid-19 detection from xray and ct scans using transfer learning

    Mohamed Berrimi, Skander Hamdi, Raoudha Yahia Cherif, Abdelouahab Moussaoui, Mourad Oussalah, and Mafaza Chabane. Covid-19 detection from xray and ct scans using transfer learning. In2021 International Conference of Women in Data Science at Taif University (WiDSTaif ), pages 1–6, 2021

    2021

  12. [20]

    A comparative study of deep learning networks for covid-19 recognition in chest x-ray images

    Sabrina Nefoussi, Abdenour Amamra, and Idir Amine Amarouche. A comparative study of deep learning networks for covid-19 recognition in chest x-ray images. In2020 2nd International Workshop on Human-Centric Smart Environments for Health and Well-being (IHSH), pages 237–241, 2021

    2021

  13. [21]

    Classification of covid-19 from chest x-ray images using deep convolutional neural network

    Sohaib Asif, Yi Wenhui, Hou Jin, and Si Jinhai. Classification of covid-19 from chest x-ray images using deep convolutional neural network. In2020 IEEE 6th International Conference on Computer and Communications (ICCC), pages 426–433, 2020

    2020

  14. [22]

    Goldgof, Rahul Paul, Dmitry B

    Kaoutar Ben Ahmed, Gregory M. Goldgof, Rahul Paul, Dmitry B. Goldgof, and Lawrence O. Hall. Discovery of a generalization gap of convolutional neural networks on covid-19 x-rays classification.IEEE Access, 9:72970– 72979, 2021

    2021

  15. [23]

    Covid-19 detection through x-ray chest images

    Diego Hernandez, Rodrigo Pereira, and Petia Georgevia. Covid-19 detection through x-ray chest images. In2020 International Conference Automatics and Informatics (ICAI), pages 1–5, 2020

    2020

  16. [24]

    Lime-enabled investigation of convolutional neural network performances in covid-19 chest x-ray detection

    Eduardo Gasca Cervantes and Wai-Yip Chan. Lime-enabled investigation of convolutional neural network performances in covid-19 chest x-ray detection. In2021 IEEE Canadian Conference on Electrical and Computer Engineering (CCECE), pages 1–6, 2021

    2021

  17. [25]

    Smote: synthetic minority over-sampling technique.Journal of artificial intelligence research, 16:321–357, 2002

    Nitesh V Chawla, Kevin W Bowyer, Lawrence O Hall, and W Philip Kegelmeyer. Smote: synthetic minority over-sampling technique.Journal of artificial intelligence research, 16:321–357, 2002

    2002

  18. [26]

    Guillaume Lemaître, Fernando Nogueira, and Christos K. Aridas. Imbalanced-learn: A python toolbox to tackle the curse of imbalanced datasets in machine learning.Journal of Machine Learning Research, 18(17):1–5, 2017

    2017

  19. [27]

    why should i trust you?

    Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. " why should i trust you?" explaining the predictions of any classifier. InProceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, pages 1135–1144, 2016. 11

    2016

Pith tools

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