COVID-Net: a tailored deep convolutional neural network design for detection of COVID-19 cases from chest X-ray images

Linda WangZhong Qiu LinAlexander Wong

article2020Scientific Reports2,770 citations

Introduces COVID-Net, an open-source deep convolutional neural network architecture tailored for COVID-19 detection from chest radiographs, accompanied by the COVIDx benchmark dataset and an explainability framework to audit model predictions.

Listen

The article addresses the urgent need for rapid, accessible screening tools during the COVID-19 pandemic, where standard RT-PCR testing faces supply shortages, variable sensitivity, and processing delays. Chest X-ray imaging offers advantages in speed, availability, and portability, yet requires expert interpretation that can create bottlenecks in high-volume settings.

The article set out to create and evaluate an open-source deep convolutional neural network called COVID-Net for detecting COVID-19 cases from chest X-ray images, along with a supporting benchmark dataset.

Researchers built the COVIDx dataset by combining and modifying five public repositories, resulting in 13,975 images from 13,870 patient cases, including the largest number of publicly available COVID-19 positive cases at the time. They employed a human-machine collaborative design process that combined residual architecture principles with machine-driven exploration via generative synthesis to produce a lightweight network tailored to the task. The model was pretrained on ImageNet, trained on COVIDx, and compared against VGG-19 and ResNet-50. An explainability method was applied to audit decision-making.

COVID-Net achieved 93.3 percent test accuracy, 91 percent sensitivity for COVID-19 cases, and 98.9 percent positive predictive value, outperforming the comparator models while requiring substantially fewer parameters and multiply-accumulate operations. It correctly focused on lung regions consistent with known clinical indicators such as ground-glass opacities. The architecture incorporated a novel lightweight projection-expansion-projection-extension pattern and selective long-range connectivity that contributed to its efficiency and performance.

These results indicate that tailored deep learning models can support faster triage and reduce the burden on radiologists and testing resources, particularly in constrained environments. The open-source release enables broader community development and reproducibility. Because the model is not production-ready, further data collection and refinement are required before clinical deployment.

The authors recommend expanding the dataset as new cases become available, improving sensitivity and positive predictive value, and extending the approach to risk stratification and hospitalization prediction. Main limitations include the relatively small number of COVID-19 cases in the current dataset and the snapshot nature of an evolving collection; results should be interpreted with caution until validated on larger, more diverse data.

DarwinAI University of Waterloo convolutional neural networks deep learning open access COVID-Net architecture deep convolutional neural network COVIDx dataset

arXiv: 2003.09871 lindawangg/COVID-Net

Paper: Deep Convolutional Neural Networks for Computer-Aided Detection: CNN Architectures, Dataset Characteristics and Transfer Learning, Hoo-Chang Shin et al. (2016). Its analysis of CNN design and ImageNet transfer learning clarifies the pretrained-network strategy that COVID-Net adapts for chest X-ray classification.
Paper: ChestX-ray8: Hospital-scale Chest X-ray Database and Benchmarks on Weakly-Supervised Classification and Localization of Common Thorax Diseases, Xiaosong Wang et al. (2017). ChestX-ray8 provides an early large-scale chest X-ray dataset and benchmark precedent that helps contextualize COVID-Net’s construction of the COVIDx resource.

Paper: Multi-Granularity Cross-modal Alignment for Generalized Medical Visual Representation Learning, Fuying Wang et al. (2022). It advances chest X-ray diagnosis beyond task-tailored classifiers by learning report-aligned representations that improve COVID-19 detection even with very few labeled examples.
Paper: ACPL: Anti-curriculum Pseudo-labelling for Semi-supervised Medical Image Classification, Fengbei Liu et al. (2022). It extends limited-label medical image classification with an imbalance-aware semi-supervised method, addressing a key challenge in COVID-Net’s small, skewed COVID-19 dataset.

Cover for COVID-Net: a tailored deep convolutional neural network design for detection of COVID-19 cases from chest X-ray images

Abstract

The Coronavirus Disease 2019 (COVID-19) pandemic continues to have a devastating effect on the health and well-being of the global population. A critical step in the fight against COVID-19 is effective screening of infected patients, with one of the key screening approaches being radiology examination using chest radiography. It was found in early studies that patients present abnormalities in chest radiography images that are characteristic of those infected with COVID-19. Motivated by this and inspired by the open source efforts of the research community, in this study we introduce COVID-Net, a deep convolutional neural network design tailored for the detection of COVID-19 cases from chest X-ray (CXR) images that is open source and available to the general public. To the best of the authors’ knowledge, COVID-Net is one of the first open source network designs for COVID-19 detection from CXR images at the time of initial release. We also introduce COVIDx, an open access benchmark dataset that we generated comprising of 13,975 CXR images across 13,870 patient patient cases, with the largest number of publicly available COVID-19 positive cases to the best of the authors’ knowledge. Furthermore, we investigate how COVID-Net makes predictions using an explainability method in an attempt to not only gain deeper insights into critical factors associated with COVID cases, which can aid clinicians in improved screening, but also audit COVID-Net in a responsible and transparent manner to validate that it is making decisions based on relevant information from the CXR images. By no means a production-ready solution, the hope is that the open access COVID-Net, along with the description on constructing the open source COVIDx dataset, will be leveraged and build upon by both researchers and citizen data scientists alike to accelerate the development of highly accurate yet practical deep learning solutions for detecting COVID-19 cases and accelerate treatment of those who need it the most.

Topics

convolutional neural networks deep learning open access high resolution image detection algorithm

Keywords

COVID-Net architecture deep convolutional neural network COVIDx dataset COVID-19 case detection chest radiography

1 Introduction
2 Related Work
3 Methods
3.1 COVIDx Dataset
3.2 Principled network design prototyping
3.3 Machine-driven Design Exploration
3.4 COVID-Net Network Architecture
3.4.1 Lightweight design pattern
3.4.2 Selective long-range connectivity
3.4.3 Architectural diversity
3.5 Implementation Details
3.6 COVID-Net Auditing via Explainability
4 Experimental Results
4.1 Quantitative Analysis
4.2 Architecture Comparisons
4.3 Qualitative Analysis
5 Conclusion
References

Knowls

Knowl 1 — COVID-Net Convolutional Neural Network Architecture
model/method
COVID-Net is a specialized convolutional neural network architecture tailored for detecting COVID-19 from chest X-ray (CXR) images, designed via a human-machine collaborative strategy. The network takes an input CXR image of dimension $480 \times 480 \times 3$ and processes it through a series of specialized stages to output predictions for three classes: normal (no infection), non-COVID-19 pneumonia, and COVID-19 viral infection.

The architecture incorporates three key design traits:
1. Projection-Expansion-Projection-Extension (PEPX) Modules: The core processing blocks utilize a multi-stage lightweight residual PEPX design comprising successive $1 \times 1$ and depthwise $3 \times 3$ convolutions, which enhances representational capacity while controlling parameter count and floating-point computations.
2. Selective Long-Range Connectivity: Instead of dense connections across all layers (which incur high memory and compute overhead), COVID-Net introduces four central $1 \times 1$ convolution hubs that route long-range connections from early processing stages directly to deep layers.
3. Architectural Diversity: The network features a heterogeneous mix of convolutional layers varying in kernel sizes ( $7 \times 7$ down to $1 \times 1$ ) and grouping configurations (standard ungrouped and depthwise convolutions) across different stages.
The sequence of feature processing consists of an initial $7 \times 7$ convolution layer ( $240 \times 240 \times 56$ ), followed by successive stages of PEPX modules (PEPX 1.1–1.3 at $120 \times 120 \times 56$ , PEPX 2.1–2.4 at $60 \times 60 \times 112$ , PEPX 3.1–3.6 at $30 \times 30 \times 216$ / $224$ , and PEPX 4.1–4.3 at $15 \times 15 \times 424$ / $400$ ), which flatten to 46,080 dimensions, pass through a 3-unit fully connected layer, and terminate in a Softmax activation.
Knowl 2 — Projection-Expansion-Projection-Extension (PEPX) Design Pattern
model/method
The Projection-Expansion-Projection-Extension (PEPX) module is a lightweight residual building block discovered through machine-driven generative synthesis. It replaces standard dense convolutional blocks to reduce computational complexity while retaining high representational capacity.

A PEPX module consists of five sequential convolutional operations:
1. First-Stage Projection: A $1 \times 1$ convolution that projects input feature channels down to a lower-dimensional bottleneck representation.
2. Expansion: A $1 \times 1$ convolution that expands the bottleneck features to a higher channel dimension differing from the input dimensionality.
3. Depth-Wise Convolution: A $3 \times 3$ depth-wise separable convolution applied channel-wise to learn spatial features with low computational overhead.
4. Second-Stage Projection: A $1 \times 1$ convolution that projects the spatial feature maps back into a lower-dimensional channel space.
5. Extension: A final $1 \times 1$ convolution that scales channel dimensionality back up to the desired target channel depth.
A residual connection adds the original input tensor to the output of the extension step where spatial dimensions match.
Knowl 3 — Human-Machine Collaborative Network Design via Generative Synthesis
model/method
COVID-Net was synthesized using a two-stage human-machine collaborative design strategy:
1. Principled Human Design Prototyping: Human designers establish an initial prototype based on deep residual architecture design principles. The prototype is structured to produce a 3-class categorical output: normal, non-COVID-19 infection (bacterial/viral), and COVID-19 viral infection.
2. Machine-Driven Generative Synthesis: The initial prototype, the training dataset, and quantitative operational constraints are provided to an exploration framework based on a generator-inquisitor pair $\{G, I\}$ . The generator $G(s; \theta_G)$ , parameterized by $\theta_G$ , generates deep neural networks $N_s = G(s)$ given seeds $s \in S$ . The inquisitor $I(G; \theta_I)$ probes network responses to targeted inputs, updates its parameters $\theta_I$ via performance metrics and response signals, and outputs parameter updates $\Delta \theta_G = I(G)$ that guide the generator toward optimal macroarchitecture and microarchitecture topologies.
The generative synthesis exploration was constrained to satisfy two explicit clinical performance requirements:
- COVID-19 sensitivity $\ge 80\%$
- COVID-19 positive predictive value (PPV) $\ge 80\%$
Knowl 4 — COVIDx Chest X-Ray Benchmark Dataset
experimental setup
The COVIDx benchmark dataset is a curated collection of chest X-ray (CXR) images constructed by aggregating and standardizing five open-access repositories: the COVID-19 Image Data Collection, the Figure 1 COVID-19 Chest X-ray Dataset Initiative, the ActualMed COVID-19 Chest X-ray Dataset Initiative, the RSNA Pneumonia Detection Challenge dataset, and the COVID-19 Radiography Database.

COVIDx contains a total of 13,975 CXR images corresponding to 13,870 patient cases across three diagnostic classes:
- Normal (no infection): 8,066 patient cases
- Non-COVID-19 Pneumonia (bacterial and non-SARS-CoV-2 viral): 5,538 patient cases
- COVID-19 Viral Infection: 266 patient cases across 358 CXR images
All images are formatted for three-class classification to assist clinical decision-making in prioritizing RT-PCR testing confirmation and distinguishing viral from bacterial treatment pathways.

Knowl 5 — Classification Accuracy and Computational Efficiency of COVID-Net

data/table

COVID-Net was evaluated on the COVIDx test dataset against two baseline deep architectures: VGG-19 and ResNet-50. COVID-Net achieved the highest overall test accuracy while significantly reducing model parameters and computational complexity (measured in multiply-accumulate operations, MACs).

COVID-Net achieved an overall accuracy of 93.3%, which is 10.3 percentage points higher than VGG-19 and 2.7 percentage points higher than ResNet-50. COVID-Net requires 11.75 million parameters (a $\sim 53\%$ reduction relative to ResNet-50) and 7.50 G MACs (a $\sim 12\times$ compute reduction compared to VGG-19 and $\sim 2.37\times$ reduction compared to ResNet-50).

Knowl 6 — Infection-Specific Sensitivity and Positive Predictive Value

data/table

The performance of COVID-Net and baseline models across individual diagnostic categories on the COVIDx test dataset was evaluated using Sensitivity and Positive Predictive Value (PPV):

	Sensitivity (%)			Positive Predictive Value (%)
Architecture	Normal	Non-COVID19	COVID-19	Normal	Non-COVID19	COVID-19
VGG-19	98.0	90.0	58.7	83.1	75.0	98.4
ResNet-50	97.0	92.0	83.0	88.2	86.8	98.8
COVID-Net	95.0	94.0	91.0	90.5	91.3	98.9

COVID-Net achieved 91.0% sensitivity for COVID-19 cases, representing an improvement of 32.3 percentage points over VGG-19 (58.7%) and 8.0 percentage points over ResNet-50 (83.0%). For positive predictive value (PPV), COVID-Net achieved 98.9% on COVID-19 cases, 90.5% on normal cases, and 91.3% on non-COVID-19 pneumonia cases. The high COVID-19 PPV limits false-positive detections, avoiding unnecessary healthcare burdens.

Knowl 7 — GSInquire Explainability Framework for Decision Auditing
model/method
GSInquire is an explainability method coupled with generative synthesis used to audit the predictions of deep neural networks. It determines the critical spatial factors in an input image that drive the classification decision of a reference network $N_{\text{ref}} = \{V_{\text{ref}}, E_{\text{ref}}\}$ , where $V$ represents network vertices and $E$ represents edges.

The framework initializes a generator $G(s; \theta_G)$ and an inquisitor $I(G; \theta_I)$ such that the generated network matches the reference network $N_s = \{V_s, E_s\} = \{V_{\text{ref}}, E_{\text{ref}}\}$ using an indicator function $1_r(\cdot)$ and a universal performance metric $U$ . Given an input stimulus signal $x$ (a CXR image), the inquisitor $I$ probes subgraphs $\{V_s', E_s'\}$ (where $V_s' \subseteq V_s$ and $E_s' \subseteq E_s$ ) and observes reactionary response signals $y \in Y_G(s)$ .

The inquisitor parameters $\theta_I$ are updated based on $Y_G(s)$ , $U(G(s))$ , and $1_r(G(s))$ . The inquisitor then outputs parameter variations $\Delta \theta_G = I(G)$ , which are mapped back to the input spatial domain via a projection transformation $T(\Delta \theta_G(s))$ to yield an interpretation map $z(x; N_{\text{ref}})$ . This spatial attribution highlights the critical pixels driving the prediction.
Knowl 8 — COVID-Net Preprocessing and Training Protocol
experimental setup
COVID-Net training is conducted using the following data preparation and optimization protocol:
1. Data Preprocessing: The top 8% of each chest X-ray image is cropped out prior to training to eliminate embedded textual labels, hospital annotations, and non-anatomical artifacts.
2. Data Augmentation: Training samples undergo random transformations, including horizontal flipping, horizontal and vertical translations (within $\pm 10\%$ ), rotations (within $\pm 10^\circ$ ), zooming (within $\pm 15\%$ ), and intensity shifts (within $\pm 10\%$ ).
3. Training Configuration: COVID-Net weights are initialized via pretraining on ImageNet. Optimization is performed using the Adam optimizer with an initial learning rate of $2 \times 10^{-4}$ , batch size of 64, and 22 training epochs.
4. Learning Rate Policy & Rebalancing: Learning rate decay is governed by a reduce-on-plateau policy with a reduction factor of 0.7 and a patience parameter of 5 epochs. A batch rebalancing strategy is applied during sampling to maintain an even distribution of classes across each mini-batch.
Knowl 9 — Explainability-Driven Validation of Radiographical Biomarkers
empirical result
An audit of COVID-Net using GSInquire confirmed that the network bases its predictions on genuine pulmonary radiographical indicators rather than extraneous visual artifacts (such as embedded text markers, patient positioning boundaries, or radiographic acquisition artifacts).

Spatial attribution maps generated for true-positive COVID-19 predictions demonstrated that the critical decision factors localized heavily within the lung fields. Specifically, the features identified by the model aligned directly with clinical visual indicators recognized by radiologists in COVID-19 patient chest radiographs, including ground-glass opacities, bilateral abnormalities, and interstitial abnormalities.

Coverage note — None was omitted; all key architectural components, dataset descriptions, mathematical explainability formulations, quantitative results, and qualitative findings were captured.

References

1.Wang, W. et al. Detection of SARS-CoV-2 in different types of clinical specimens. JAMA 323(18), 1843–1844 (2020).
2.West, C. P., Montori, V. M., & Sampathkumar, P. Covid-19 testing: the threat of false-negative results. In Mayo Clinic Proceeding (2020).
3.Fang, Y. et al. Sensitivity of chest CT for covid-19: comparison to RT-PCR. Radiology. https://doi.org/10.1148/radiol.2020200432 (2020).
4.Yang, Y. et al. Evaluating the accuracy of different respiratory specimens in the laboratory diagnosis and monitoring the viral shedding of 2019-ncov infections. medRxiv:2020.02.11.20021493 (2020).
5.Wikramaratna, P., Paton, R. S., Ghafari, M., & Lourenco, J. Estimating false-negative detection rate of sars-cov-2 by rt-pcr. medRxiv:2020.04.05.20053355 (2020).
6.Ng, M.-Y. et al. Imaging profile of the COVID-19 infection: radiologic findings and literature review. Radiol. Cardiothorac. Imaging 2(1), e200034 (2020).
7.Huang, C. et al. Clinical features of patients infected with 2019 Novel Coronavirus in Wuhan China. The Lancet 395, 497–506 (2020).
8.Guan, W. J., Hu Y., & Ni Z. Y. Clinical characteristics of Coronavirus disease 2019 in China. N. Engl. J. Med. 382(18), 1708–1720 (2020).
9.Ai, T. et al. Correlation of chest CT and RT-PCR testing in coronavirus disease 2019 (COVID-19) in china: a report of 1014 cases. Radiology (2020). https://doi.org/10.1148/radiol.2020200642.
10.Rubin, G. D. et al. The role of chest imaging in patient management during the COVID-19 pandemic: a multinational consensus statement from the fleischner society. Radiology (2020). https://doi.org/10.1016/j.chest.2020.04.003.
11.Nair, A. et al. A British Society of thoracic imaging statement: considerations in designing local imaging diagnostic algorithms for the COVID-19 pandemic. Clin. Radiol. 75(5), 329–334 (2020).
12.Dennie, C. et al. The Canadian Society of Thoracic Radiology (CSTR) and Canadian Association of Radiologists (CAR) consensus statement regarding chest imaging in suspected and confirmed COVID-19. Can. Assoc. Radiol. J. 71(4), 470–481 (2020).
13.Jacobi, A., Chung, M., Bernheim, A. & Eber, C. Portable chest x-ray in coronavirus disease-19 (covid-19): a pictorial review. Clin. Imaginghttps://doi.org/10.1016/j.clinimag.2020.04.001 (2020).
14.Wu, G., & Li, X. Mobile x-rays are highly valuable for critically ill covid patients. Eur. Radiol.. https://doi.org/10.1007/s00330-020-06918-2(2020).
15.Mao, B. et al. Assessing risk factors for SARS-CoV-2 infection in patients presenting with symptoms in Shanghai, China: a multicentre observational cohort study. Lancet Dig. Health. https://doi.org/10.1016/S2589-7500(20)30109-6(2020).
16.Cohen, J. P., Morrison, P., & Dao, L. COVID-19 image data collection. arXiv:2003.11597 (2020).
17.Chung, A. Figure 1 COVID-19 chest x-ray data initiative. https://github.com/agchung/Figure1-COVID-chestxray-dataset (2020).
18.Chung, A. Actualmed COVID-19 chest x-ray data initiative. https://github.com/agchung/Actualmed-COVID-chestxray-dataset (2020).
19.Radiological Society of North America. COVID-19 radiography database. https://www.kaggle.com/tawsifurrahman/covid19-radiography-database (2019).
20.Radiological Society of North America. RSNA pneumonia detection challenge. https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/data (2019).
21.LeCun, Y., Bengio, Y., & Hinton, G. Deep learning. Nature 521, 436–444 (2015).
22.Gozes, O. et al. Rapid AI development cycle for the coronavirus (COVID-19) pandemic: initial results for automated detection and patient monitoring using deep learning ct image analysis. arXiv:2003.05037 (2020).
23.Xu, X. et al. Deep learning system to screen coronavirus disease 2019 pneumonia. arXiv:2002.09334 (2020).
24.Li, L. et al. Artificial intelligence distinguishes COVID-19 from community acquired pneumonia on chest CT. Radiology. https://doi.org/10.1148/radiol.2020200905 (2020).
25.Shi, F. et al. Large-scale screening of covid-19 from community acquired pneumonia using infection size-aware classification. arXiv:2003.09860 (2020).
26.Li, X., Li, C., & Zhu, D. Covid-mobilexpert: On-device covid-19 screening using snapshots of chest x-ray. arXiv:2004.03042 (2020).
27.Minaee, S., Kafieh, R., Sonka, M., Yazdani, S., & Soufi, G. J. Deep-covid: predicting covid-19 from chest x-ray images using deep transfer learning. arXiv:2004.09363 (2020).
28.Afshar, P. et al. Covid-caps: a capsule network-based framework for identification of covid-19 cases from x-ray images. arXiv:2004.02696 (2020).
29.Luz, E. et al. Towards an effective and efficient deep learning model for covid-19 patterns detection in x-ray images. arXiv:2004.05717 (2020).
30.Khobahi, S., Agarwal, C., & Soltanalian, M. Coronet: a deep network architecture for semi-supervised task-based identification of covid-19 from chest x-ray images. medRxiv. https://doi.org/10.1101/2020.04.14.20065722 (2020).
31.Ucar, F., & Korkmaz, D. COVIDiagnosis-net: deep bayes-squeezenet based diagnosis of the coronavirus disease 2019 (COVID-19) from x-ray images. Med. Hypotheses. https://doi.org/10.1016/j.mehy.2020.109761 (2020).
32.Tartaglione, E., Barbano, C. A., Berzovini, C., Calandri, M., & Grangetto, M. Unveiling covid-19 from chest x-ray with deep learning: a hurdles race with small data. arXiv:2004.05405 (2020).
33.Yeh, C.-F. et al. A cascaded learning strategy for robust covid-19 pneumonia chest x-ray screening. arXiv:2004.12786 (2020).
34.Zhang, Y. et al. Covid-da: deep domain adaptation from typical pneumonia to covid-19. arXiv:2005.01577 (2020).
35.Karim, M. R. et al. Deepcovidexplainer: explainable covid-19 predictions based on chest x-ray images (2020).
36.Apostolopoulos, I. D., & Mpesiana, T. A. Covid-19: automatic detection from x-ray images utilizing transfer learning with convolutional neural networks. Phys. Eng. Sci. Med. 43, 635–640. https://doi.org/10.1007/s13246-020-00865-4 (2020).
37.Farooq, M., & Hafeez, A. Covid-resnet: a deep learning framework for screening of covid19 from radiographs (2020).
38.Majeed, T., Rashid, R., Ali, D., & Asaad, A. Covid-19 detection using cnn transfer learning from x-ray images. medRxiv (2020).
39.Wang, X. et al. Chestx-ray8: hospital-scale chest x-ray database and benchmarks on weakly-supervised classification and localization of common thorax diseases. In 2017 IEEE Conference on Computer Vision and Pattern Recognition(CVPR), 3462–3471 (2017).
40.He, K., Zhang, X., Ren, S., & Sun, J. Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition 770–778 (2016).
41.Wong, A., Shafiee, M. J., Chwyl, B., & Li, F. Ferminets: learning generative machines to generate efficient neural networks via generative synthesis. arXiv preprint arXiv:1809.05989 (2018).
42.Huang, G., Liu, Z., & Weinberger, K. Q. Densely connected convolutional networks. arXiv preprint arXiv:1608.06993 (2016).
43.Deng, J. et al. Imagenet: a large-scale hierarchical image database. In 2009 IEEE Conference on Computer Vision and Pattern Recognition 248–255 (IEEE, 2009).
44.Lin, Z. Q. et al. Do explanations reflect decisions? A machine-centric strategy to quantify the performance of explainability algorithms. arXiv preprint arXiv:1910.07387 (2019).
45.Wong, A. Netscore: towards universal metrics for large-scale performance analysis of deep neural networks for practical usage. CoRR. arXiv preprint arXiv:abs/1806.05512 (2018).
46.Simonyan, K., & Zisserman, A. Very deep convolutional networks for large-scale image recognition (2014).

Citation

MLA

Wang, L., and A. Wong. “COVID-Net: A Tailored Deep Convolutional Neural Network Design for Detection of COVID-19 Cases from Chest X-Ray Images”. arXiv, 2020, http://arxiv.org/abs/2003.09871v4.

APA

Wang, L., & Wong, A. (2020). COVID-Net: A Tailored Deep Convolutional Neural Network Design for Detection of COVID-19 Cases from Chest X-Ray Images. arXiv. http://arxiv.org/abs/2003.09871v4

Chicago

Wang, L., and A. Wong. 2020. “COVID-Net: A Tailored Deep Convolutional Neural Network Design for Detection of COVID-19 Cases from Chest X-Ray Images”. arXiv. http://arxiv.org/abs/2003.09871v4.

Harvard

Wang, L. and Wong, A. (2020) “COVID-Net: A Tailored Deep Convolutional Neural Network Design for Detection of COVID-19 Cases from Chest X-Ray Images”, arXiv [Preprint]. Available at: http://arxiv.org/abs/2003.09871v4.

Vancouver

1. Wang L, Wong A (2020) COVID-Net: A Tailored Deep Convolutional Neural Network Design for Detection of COVID-19 Cases from Chest X-Ray Images. arXiv

BibTeX

@article{wang2020covid,
  title = {COVID-Net: A Tailored Deep Convolutional Neural Network Design for Detection of COVID-19 Cases from Chest X-Ray Images},
  author = {Wang, Linda and Wong, Alexander},
  year = {2020},
  journal = {arXiv},
  url = {http://arxiv.org/abs/2003.09871v4},
  eprint = {2003.09871}
}

Metadata:arXiv

Source Code

This paper has an official code repository available. Click below to access the source code.

View Repository

Access the Paper

This paper is available from its original source. Click below to access the PDF.

Open PDF

Architecture	Params (M)	MACs (G)	Acc. (%)
VGG-19	20.37	89.63	83.0
ResNet-50	24.97	17.75	90.6
COVID-Net	11.75	7.50	93.3

COVID-Net: a tailored deep convolutional neural network design for detection of COVID-19 cases from chest X-ray images

Abstract

Topics

Keywords

Table of Contents

Knowls

Knowl 1 — COVID-Net Convolutional Neural Network Architecture

Knowl 2 — Projection-Expansion-Projection-Extension (PEPX) Design Pattern

Knowl 3 — Human-Machine Collaborative Network Design via Generative Synthesis

Knowl 4 — COVIDx Chest X-Ray Benchmark Dataset