CoChat
Sign Up
Cited by 1,913COVID-19 diagnosis using AIArtificial Intelligence in Healthcare and EducationSARS-CoV-2 detection and testing

Can AI help in screening viral and COVID-19 pneumonia?

Open access
Chowdhury, Muhammad E. H., Rahman, Tawsifur, Khandakar, Amith, Mazhar, Rashid, Kadir, Muhammad Abdul, Mahbub, Zaid Bin, Islam, Khandakar Reajul, Khan, Muhammad Salman +4 more·2020·IEEE Access
DOI 10.1109/access.2020.3010287
Open this paper in CoChat — free accountOpen PDF

Abstract

Coronavirus disease (COVID-19) is a pandemic disease, which has already caused thousands of causalities and infected several millions of people worldwide. Any technological tool enabling rapid screening of the COVID-19 infection with high accuracy can be crucially helpful to the healthcare professionals. The main clinical tool currently in use for the diagnosis of COVID-19 is the Reverse transcription polymerase chain reaction (RT-PCR), which is expensive, less-sensitive and requires specialized medical personnel. X-ray imaging is an easily accessible tool that can be an excellent alternative in the COVID-19 diagnosis. This research was taken to investigate the utility of artificial intelligence (AI) in the rapid and accurate detection of COVID-19 from chest X-ray images. The aim of this paper is to propose a robust technique for automatic detection of COVID-19 pneumonia from digital chest X-ray images applying pre-trained deep-learning algorithms while maximizing the detection accuracy. A public database was created by the authors combining several public databases and also by collecting images from recently published articles. The database contains a mixture of 423 COVID-19, 1485 viral pneumonia, and 1579 normal chest X-ray images. Transfer learning technique was used with the help of image augmentation to train and validate several pre-trained deep Convolutional Neural Networks (CNNs). The networks were trained to classify two different schemes: i) normal and COVID-19 pneumonia; ii) normal, viral and COVID-19 pneumonia with and without image augmentation. The classification accuracy, precision, sensitivity, and specificity for both the schemes were 99.7%, 99.7%, 99.7% and 99.55% and 97.9%, 97.95%, 97.9%, and 98.8%, respectively. The high accuracy of this computer-aided diagnostic tool can significantly improve the speed and accuracy of COVID-19 diagnosis. This would be extremely useful in this pandemic where disease burden and need for preventive measures are at odds with available resources.

AI snapshot

Written from the paper’s full text.

TL;DR

This paper proposes transfer learning with pre-trained CNNs to detect COVID-19 pneumonia from chest X-ray images, using a newly assembled public database of 423 COVID-19, 1,485 viral pneumonia, and 1,579 normal images. It reports very high classification performance, especially for DenseNet201 with image augmentation in the three-class task.

Key findings

  • For two-class classification (normal vs. COVID-19 pneumonia), the reported accuracy, precision, sensitivity, and specificity were 99.7%, 99.7%, 99.7%, and 99.55%, respectively.
  • For three-class classification (normal, viral pneumonia, and COVID-19 pneumonia), the reported accuracy, precision, sensitivity, and specificity were 97.9%, 97.95%, 97.9%, and 98.8%, respectively.
  • CheXNet achieved the highest two-class accuracy of 99.4% without augmentation and 99.7% with augmentation; DenseNet201 outperformed other models in the three-class scheme when image augmentation was used.
  • In the DenseNet201 confusion matrix with augmentation, only 3 of 423 COVID-19 images were misclassified as normal in the two-class problem, and only 1 COVID-19 image was misclassified as normal in the three-class problem.

Methods

  • The authors created a public database by combining several public sources and collecting images from published articles: 423 COVID-19, 1,485 viral pneumonia, and 1,579 normal chest X-ray images.
  • Eight pre-trained CNNs were trained, validated, and tested using transfer learning: MobileNetv2, SqueezeNet, ResNet18, ResNet101, DenseNet201, CheXNet, Inceptionv3, and VGG19.
  • Two classification schemes were evaluated with and without image augmentation: two-class (normal vs. COVID-19) and three-class (normal, viral pneumonia, and COVID-19), using stratified 5-fold cross-validation with 80% training and 20% test splits.

Limitations

  • The paper does not include a dedicated limitations section; it states that three COVID-19 images were misclassified and that cases without evident light-focus edge features in deep layers “have to be confirmed by other techniques.”
  • The paper notes that results reported on smaller databases in the literature should be evaluated on a large dataset to be generalizable.
  • Evident from the text: the study uses a retrospectively assembled public/image-article dataset rather than a prospective clinical cohort, and no external validation cohort or full reader study is described; only three misclassified images were evaluated by three practicing radiologists.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1109/access.2020.3010287

Cites

64 of 64 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 21 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    The COVID-19 sub-database comprises 423 AP/PA images created from four major data sources.

    “COVID-19 sub-database, comprising of 423 AP/PA images, was created from the following four major data sources.”
    Quote in the paperSubstantial claim
  • Findingnone given

    The normal and viral pneumonia sub-databases of 1579 and 1485 X-ray images were created from the RSNA-Pneumonia-Detection-Challenge database and the Kaggle Chest X-ray Images database.

    “Images from RSNA-Pneumonia-Detection-Challenge database along with the Chest X-ray Images database from Kaggle were used to create the normal and viral pneumonia sub-databases of 1579 and 1485 X-ray images respectively.”
    Quote in the paperSubstantial claim
  • Findingown data

    All evaluated pre-trained models perform very well in classifying COVID-19 and normal images in the two-class problem.

    “It is apparent from Table 2 that all the evaluated pre-trained models perform very well in classifying COVID-19 and normal images in two-class problem.”
    Quote in the paperSubstantial claim
  • Findingreasoning

    When image augmentation is applied to the training image set, all pre-trained networks perform based on their capability to distinguish the three-class images.

    “On the other hand, while the image augmentation was applied to the training image set, all the pre-trained networks are now performing based on their capability to distinguish the three-class images.”
    Quote in the paperNot a substantial claim
  • Findingown data

    The 14th layer of DenseNet201 can detect features that can distinguish normal, COVID-19 and viral pneumonia images.

    “It is notable from Figure 8 that the 14 th layer of the DenseNet201 can detect features that can distinguish normal, COVID-19 and Viral Pneumonia images.”
    Quote in the paperSubstantial claim
  • May combine two claims. 34
  • The quote only partly states this claim. 1
  • Classified as probably not a substantial claim. 11

  • May combine two claims.

    Findingown data

    The authors prepared a large database of X-ray images of 1579 normal, 1485 viral pneumonia, and 423 COVID-19 positive pneumonia and made it publicly available.

    “The authors in this paper have prepared a large database of X-ray images of 1579 normal, 1485 viral pneumonia and 423 COVID-19 positive pneumonia and made this publicly available so that other researchers can get benefit from it.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingcitation

    The authors collected 60 COVID-19 positive chest X-ray images from 43 recently published articles and 32 positive chest X-ray images from Radiopaedia, which were not listed in the GitHub database.

    “Authors managed to collect 60 COVID-19 positive chest X-ray images from 43 recently published articles [60], which were not listed in the GitHub database and 32 positive chest x-ray images from Radiopaedia [63], which were not listed in the GitHub database.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [60] source

      Not checked in the free run.

    • cites [63]

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    CheXNet produced the highest accuracy of 99.4% without augmentation and 99.7% with augmentation for two-class classification.

    “CheXNet is producing the highest accuracy of 99.4% and 99.7% for two-class classification without and with image augmentation respectively.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Among networks trained with 338 X-ray images for the two-class problem, ResNet18 and CheXNet perform equally, while CheXNet and DenseNet201 perform better than others with augmented images, although the difference is marginal.

    “Among the networks trained with 338 X-ray images for two-class problem, ResNet18 and CheXNet are equally performing for classifying images while CheXNet and DenseNet201 are performing better than others in case of training with augmented images, although the difference is marginal.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The weighted average performance metrics for eight networks are very similar, with a small gain when training used image augmentation.

    “The weighted average performance matrix for eight different networks are very similar whereas small gain can be observed when training was done using image augmentation.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The strongest activation channel of the CNN activates on edges with positive activation on light left/dark right edges and negative activation on dark left/light right edges.

    “It was noticed that the strongest channel activates on edges with positive activation on light left/dark right edges, and negative activation on dark left/light right edges.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    In two-class classification with image augmentation, only three COVID-19 images out of 423 were miss-classified to normal (false negative) and only three images out of 1579 were miss-classified to COVID-19 (false positive).

    “It is clear from Figure 7(A) that only three COVID-19 images out of 423 images were miss-classified to normal (false negative) and only three images out of 1579 images were miss-classified to COVID-19 (false positive).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    None of the normal images were miss-classified to COVID-19 by the three-class classifier, although several normal images were miss-classified to viral pneumonia.

    “None of normal images were miss-classified to COVID-19 by the three-class classifier although several normal images were miss-classified to viral pneumonia.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In the three-class problem, only one COVID-19 image was miss-classified to normal, which is one of the three images miss-classified by the two-class classifier.

    “In the three-class problem, only one COVID-19 image was miss-classified to normal, which is one of the three images miss-classified by the two-class classifier.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingreasoning

    For the three-class problem, when pre-trained networks are trained on a small image dataset, the performance difference is very marginal and overall performance is reduced compared to the two-class problem.

    “If the pre-trained networks are trained on a small image dataset as reported by the most of the research groups in the literature, the performance difference is very marginal and overall performance is reduced for three-class problem in comparison to two-class problem.”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingown data

    DenseNet201 outperforms other models in the three-class classification scheme in terms of different performance indices when image augmentation was employed.

    “It was observed that DenseNet201 outperforms other different deep CNN networks while image augmentation was used for training the CNN models.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Only four viral pneumonia images were miss-classified to COVID-19 out of 1485 images, while 33 images were miss-classified to normal.

    “Only four viral pneumonia images were miss-classified to COVID-19 out of 1485 images while 33 images miss-classified to normal.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For the two-class scheme, the classification accuracy, precision, sensitivity, and specificity were 99.7%, 99.7%, 99.7%, and 99.55%, respectively.

    “The classification accuracy, precision, sensitivity, and specificity for both the schemes were 99.7%, 99.7%, 99.7% and 99.55% and 97.9%, 97.95%, 97.9%, and 98.8%, respectively.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For the three-class scheme, the classification accuracy, precision, sensitivity, and specificity were 97.9%, 97.95%, 97.9%, and 98.8%, respectively.

    “The classification accuracy, precision, sensitivity, and specificity for both the schemes were 99.7%, 99.7%, 99.7% and 99.55% and 97.9%, 97.95%, 97.9%, and 98.8%, respectively.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    First and third images were identified as no sign or very little sign of COVID-19 by the radiologists, while image-02 was identified as very mild stage of lung infections.

    “First and third images were identified as no sign or very little sign of COVID-19 by the radiologists while image-02 was identified as very mild stage of lung infections.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    CheXNet, a variant of DenseNet, was outperforming other networks while image augmentation was not used.

    “CheXNet which is a variant of DenseNet was outperforming other networks while image augmentation was not used.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For normal and COVID-19 classification, accuracy, precision, sensitivity, and specificity were 99.7%, 99.7%, 99.7%, and 99.55%, respectively.

    “The classification accuracy, precision, sensitivity, and specificity of normal and COVID-19 images, and normal, COVID-19 and viral pneumonia were (99.7%, 99.7%, 99.7% and 99.55%), and (97.9%, 97.95%, 97.9%, and 98.8%) respectively.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingnone given

    For normal, COVID-19, and viral pneumonia classification, accuracy, precision, sensitivity, and specificity were 97.9%, 97.95%, 97.9%, and 98.8%, respectively.

    “The classification accuracy, precision, sensitivity, and specificity of normal and COVID-19 images, and normal, COVID-19 and viral pneumonia were (99.7%, 99.7%, 99.7% and 99.55%), and (97.9%, 97.95%, 97.9%, and 98.8%) respectively.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    There is a significant shortage of RT-PCR supply, which leads to delay in disease prevention efforts.

    “Moreover, there is a significant shortage of its supply, which leads to delay in the disease prevention efforts [6].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [6] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Convolutional Neural Networks have been proven extremely beneficial in feature extraction and learning.

    “Specifically, Convolutional Neural Network (CNN) has been proven extremely beneficial in feature extraction and learning and therefore, widely adopted by the research community [26].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [26]

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    The small datasets used by recent groups make it difficult to generalize their results and cannot guarantee that reported performance will retain on a larger dataset.

    “This makes it difficult to generalize their results reported in these articles and cannot guarantee that the reported performance will retain when these models will be tested on a larger dataset.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    RT-PCR is manual, complicated, laborious, time-consuming, and has a positivity rate of only 63%.

    “RT-PCR is used as a reference method for the detection of COVID-19 patients, however, the technique is manual, complicated, laborious and time-consuming with a positivity rate of only 63% [5].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [5] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    A rigorous experiment on a large database of COVID and non-COVID classes is very few and missing in the case of transfer learning approach.

    “A rigorous experiment on a large database of COVID and non-COVID classes are very few and missing in case of transfer learning approach.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [40]- [56]

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationreasoning

    X-ray imaging is an easily accessible tool that can be an excellent alternative in COVID-19 diagnosis.

    “X-ray imaging is an easily accessible tool that can be an excellent alternative in the COVID-19 diagnosis.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    The RT-PCR kit costs about USD 120-130 and requires a specialized biosafety lab costing USD 15,000 to USD 90,000.

    “It is reported that the RT-PCR kit costs about USD 120-130 and also requires a specialized biosafety lab to house the PCR machine, each of which may cost USD 15,000 to USD 90,000 [8].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [8] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductreasoning

    In study2, image augmentation was applied to create a balanced training set, with COVID-19 images augmented six times and normal and viral pneumonia images augmented once.

    “Image augmentation techniques were applied to viral pneumonia, normal and COVID-19 X-ray images for training to create a balanced training set.However, COVID-19 images were augmented six times while normal and viral pneumonia images were augmented once only.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    Transfer learning uses a trained model from a large dataset such as ImageNet and modifies the Softmax and classification layer of the pre-trained network.

    “The concept of transfer learning uses the trained model from large dataset such as ImageNet [57] and modify the Softmax and classification layer of the pre-trained networks.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [57] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    Both experiments were evaluated using a stratified 5-fold cross-validation scheme with 80% training and 20% test splits, with 10% of training data used as a validation set to avoid overfitting.

    “Both the experiments were evaluated using a stratified 5-fold cross-validation (CV) scheme with a ratio of 80% for training and 20% for the test (unseen folds) splits, where 10% of training data is used as a validation set to avoid overfitting.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    Fivefold cross-validation results were averaged to produce the final ROC curve, confusion matrix, and evaluation matrices.

    “Fivefold cross-validation result was averaged to produce the final receiver operating characteristic (ROC) curve, confusion matrix, and evaluation matrices.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    In study1, image augmentation was not applied, and equal numbers of normal and viral pneumonia images were randomly selected to match the 423 COVID-19 images to balance the database.

    “In the study1, image augmentation technique was not applied to the training data.Since COVID-19 positive chest X-ray images were 423, same number of X-ray images were randomly selected from normal (out of 1579) and viral pneumonia (out of 1485) images to match with COVID-19 images to balance the database.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationreasoning

    CheXNet performs well with and without augmentation because it is the only network pre-trained on a large X-ray image database and is supposed to perform better for X-ray image classification without retraining on a larger dataset.

    “Interestingly, CheXNet is performing well in both the cases, with and without augmentation and this can be explained from the fact that CheXNet is the only network which is pre-trained on a large.X-ray image database and the network supposed to perform better for X-ray image classification without the requirement of training again on a larger dataset.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    In the two-class problem, because COVID-19 images are significantly different from normal images, all tested networks perform well.

    “However, in this classification problem as the COVID-19 images are significantly different from normal images all the tested networks are performing well.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: COVID-19 X-ray images are significantly different from normal images.

  • May combine two claims.

    Premisecitation

    Transfer learning removes the requirement of having a large dataset and reduces the long training period required by deep learning developed from scratch.

    “This removes the requirement of having large dataset and also reduces the long training period as is required by the deep learning algorithm when developed from scratch [58], [59].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [58]

      Not checked in the free run.

    • cites [59]

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    This tool would be extremely useful in the pandemic where disease burden and need for preventive measures are at odds with available resources.

    “This would be extremely useful in this pandemic where disease burden and need for preventive measures are at odds with available resources.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationreasoning

    Typically, the deeper the network, the better its performance in distinguishing the image classes.

    “Typically, the deeper the network the better is the performance in distinguishing the image classes.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The success of the network in detecting COVID-19 X-ray images and distinguishing it from normal and viral pneumonia images is due to deep layer features; several groups of researchers reported earlier that this is not reliably possible by plain X-ray images.

    “This shows the reason of the success of the network in detecting COVID-19 X-ray images and distinguishing it from normal and viral pneumonia images, which several groups of researchers reported earlier are not reliably possible by plain X-ray images [69]- [72].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [69]- [72]

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    Training an even deeper network with a larger image set can give a better chance of training from new image sets; deep networks can learn better and perform better if training is carried out on a larger dataset.

    “Therefore, it can be summarized that even though CheXNet was trained originally on X-ray images but training an even deeper network with a larger image set can give better chance of training from the new image sets on which the training is done, i.e., deep network can learn better and perform better if the training is carried out on a larger dataset.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The network is not confusing between COVID-19 and the other two image classes; rather, it is more confused between viral pneumonia and normal images.

    “It can be noted that network is not confusing between COVID-19, and other two image classes rather network is more confused between viral pneumonia and normal images.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Computer-aided diagnosis can save lives by early screening and proper care.

    “Since a large number of patients attending out-door or emergency, doctor's time is limited and computeraided-diagnosis can save lives by early screening and propercare.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Doctor's time is limited because a large number of patients attend out-door or emergency.

  • May combine two claims.

    Interpretationown data

    Performance reported on smaller databases in the literature should be evaluated on a large dataset; otherwise, the findings of these studies cannot be generalized for real applications.

    “This clearly reveals the fact that the performance reported on smaller database in the literature should be evaluated on a large dataset otherwise, the findings of these studies cannot be generalized for real applications.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    The algorithm fails if no evident light focus edge feature appears in the deep layer, and this type of COVID-19 cases have to be confirmed by other techniques.

    “The algorithm fails if no evident light focus edge feature is appeared in the deep layer and this type of COVID-19 cases have to be confirmed by other techniques.”
    Quote in the paperSubstantial claim
41 of 41

  • The paper relies on this without support.

    Deep convolutional neural networks typically perform better with a larger dataset than a smaller one.

    “Deep convolutional neural networks typically perform better with a larger dataset than a smaller one.”

  • The paper relies on this without support.

    The number of chest X-ray images publicly available online is small and scattered.

    “Although there are a large number of COVID-19 patients infected worldwide, the number of chest X-ray images publicly available online are small and scattered.”

  • The paper relies on this without support.

    COVID-19 X-ray images are significantly different from normal images.

    “as the COVID-19 images are significantly different from normal images all the tested networks are performing well.”

  • The paper relies on this without support.

    Images of various viral pneumonias are similar and overlap with other infectious and inflammatory lung diseases.

    “the images of various viral pneumonias are similar and they overlap with other infectious and inflammatory lung diseases.”

  • The paper relies on this without support.

    The symptoms of COVID-19 are similar to those of viral pneumonia.

    “The symptoms of COVID-19 being similar to that of viral pneumonia”

  • The paper relies on this without support.

    CheXNet was originally trained on a very large X-ray image dataset.

    “CheXNet was originally trained one a very large X-ray image dataset.”
    “CheXNet was pre-trained on a large X-ray database”

  • The paper relies on this without support.

    There is a large degree of variability in the input images from X-ray machines due to variations in the expertise of the radiologist.

    “Moreover, there is a large degree of variability in the input images from the X-ray machines due to the variations of expertise of the radiologist.”

  • The paper relies on this without support.

    The evaluations of the three practicing radiologists are reliable for identifying signs of COVID-19 in the miss-classified images.

    “These three images were evaluated by three practicing radiologists to identify what is their evaluations for these three images.”

  • The paper relies on this without support.

    Doctor's time is limited because a large number of patients attend out-door or emergency.

    “Since a large number of patients attending out-door or emergency, doctor's time is limited”
9 of 9

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • IMISNet demonstrates exceptional proficiency in CT datasets but its performance significantly decreases on X-Ray, as observed in the Radiography results.

    “For instance, while specialized models like IMISNet demonstrate exceptional proficiency in CT datasets, their performance significantly decreases when applied to X-Ray, as observed in the Radiography (Chowdhury et al., 2020) results.”

    From MedSAM-Agent: Empowering Interactive Medical Image Segmentation with Multi-turn Agentic Reinforcement Learning (2026) · cites this paper as Chowdhury et al., 2020

    The Database Description identifies the authors’ radiography dataset as chest X-ray images, supporting the reference to Radiography as an X-ray dataset. The cited paper’s experimental results concern X-ray classification; our copy does not report IMISNet’s CT performance or a decrease when it is applied to X-rays (searched: IMISNet, CT datasets and X-ray performance). (LLM, conf=0.79)

    Experimental — accuracy not yet measured

  • MedSAM-Agent outperforms ideal single-turn baselines (Single-Turn Point and Single-Turn Box) on evaluated datasets.

    “As illustrated in Fig. 3, which details the performance trajectories on the FH-PS-AOP (Jieyun & ZhanHong, 2024) and Radiography (Chowdhury et al., 2020) datasets, MedSAM-Agent outperforms these ideal single-turn baselines.”

    From MedSAM-Agent: Empowering Interactive Medical Image Segmentation with Multi-turn Agentic Reinforcement Learning (2026) · cites this paper as Chowdhury et al., 2020

  • COVID-19 Chest X-ray is a publicly-available database containing 3616 COVID-19 positive cases, 10,192 normal controls, 6012 lung opacity, and 1345 viral pneumonia cases.

    “COVID-19 Chest X-ray (also known as COVID-19 CXR) [109] 9 is a publicly-available database of chest X-ray images, containing 3616 COVID-19 positive cases, 10,192 normal controls, 6012 lung opacity (non-COVID infection), and 1345 viral pneumonia cases.”

    From Federated learning for medical image analysis: A survey (2024) · cites this paper as [109]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Citation [51] does not identify the Khan study

    Minor

    The text cites [51] for Khan et al.'s study, but reference [51] lists Ucar and Korkmaz; the Khan et al. paper is listed at [56].

  • Thread-reader dataset cites an unrelated reference

    Minor

    The dataset description cites [64] for the Spanish physician's thread-reader images, but reference [64] is a ChestX-ray8 database paper.

  • Citation [47] does not identify the Ucar study

    Minor

    The text cites [47] for Ucar et al.'s SqueezeNet work, but reference [47] is a paper by Zhang et al.; the Ucar and Korkmaz paper is listed at [51].

  • COVID-CAPS is attributed to the wrong reference

    Minor

    The text cites [43] for COVID-CAPS, whereas reference [43] is by Apostolopoulos and Mpesiana; the COVID-CAPS paper appears at [49].

  • Citation [39] does not identify the Ioannis study

    Minor

    The text attributes a COVID-19 classification study to Ioannis et al. at [39], but reference [39] is a paper by Lakhani and Sundaram on tuberculosis.

5 of 5

Cites

Flagged first, then by citations

8 of the 72 references couldn’t be matched to a known paper.

Research works printed but not linked: 18

  1. weinstock, 2020, Chest X-ray findings in 636 ambulatory patients with COVID-19 presenting to an urgent care center: A normal chest X-ray is no guarantee, J Urgent Care Med, 14, 8
  2. gershgorn, 2017, The Data that Transformed AI Research-and Possibly the World
  3. muhammad, 2020, COVID-19 Chest X-Ray Database
  4. monteral, 2020, COVID-Chestxray Database
  5. radiology, 2020, COVID-19 Database
  6. 2020, Radiopedia
  7. imaging, 2020, This is a Thread of COVID-19 CXR (All SARS-CoV-2 PCR+) From my Hospital (Spain) I Hope it Could Help
  8. 10.1109/cvpr.2017.369 · DOI · Paper page
  9. mooney, 2018, Chest X-Ray Images (Pneumonia)
  10. 2020, Inception Understanding Various Architectures of Convolutional Networks
  11. 2020, Coronavirus Disease 2019 (COVID-19
  12. 2020, WHO Director-General’s Opening Remarks at the Media Briefing on COVID-19–11 March 2020
  13. medicine, 2020, Coronavirus COVID-19 Global Cases by the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University (JHU)
  14. 2020, Global COVID-19 Report
  15. news, 2020, Bangladesh Scientists Create 3 Kit Can it Help Detect COVID-19
  16. news, 2020, India’s Poor Testing Rate May Have Masked Coronavirus Cases
  17. wetsman, 2020, Coronavirus Testing Shouldn’T Be This Complicated
  18. joaquin, 2020, Using Deep Learning to Detect Pneumonia Caused by NCOV-19 From X-Ray Images

Cited by

Sorted by citations

Related

Authors
Chowdhury, M. E. H., Rahman, T., Khandakar, A., Mazhar, R., Kadir, M. A., Mahbub, Z. B., Islam, K. R., Khan, M. S., Iqbal, A., Emadi, N. A., Reaz, M. B. I., & Islam, M. T.
Affiliations
  • Qatar University (QA)
  • University of Dhaka (BD)
  • Hamad General Hospital (QA)
  • North South University (BD)
  • University of Engineering and Technology Peshawar (PK)
  • National University of Malaysia (MY)
ORCID
Venue
IEEE Access
Publisher
Institute of Electrical and Electronics Engineers (IEEE)
Published
2020
Volume
8
Pages
132665-132676
ISSN
2169-3536 (ISSN-L)
DOI
10.1109/access.2020.3010287
OpenAlex
W3013277995
Cited by
1,913
References
72
Citations by year
2026: 122 · 2025: 188 · 2024: 276 · 2023: 364 · 2022: 432 · 2021: 424 · 2020: 107
FWCI
161.58
Type
journal article
Language
en
Open access
Licence
Locations
  • IEEE Access (publishedVersion, cc-by)
  • arXiv (Cornell University) (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Figshare (submittedVersion, cc-by)
  • Qatar University QSpace (Qatar University) (submittedVersion)
  • arXiv (Cornell University)
Funders
Qatar National Research Fund, a member of Qatar Foundation, Doha, Qatar (NPRP12S-0227-190164)
Topics
  • Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › COVID-19 diagnosis using AI
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Health Sciences › Medicine › Infectious Diseases › SARS-CoV-2 detection and testing
Keywords
COVID-19 pneumonia, viral pneumonia, chest radiography, computer-aided diagnosis, COVID-19 screening, transfer learning, convolutional neural networks, image augmentation, medical image classification, COVID-19 diagnosis

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review