[Paper Review] COVID-Net: A Tailored Deep Convolutional Neural Network Design for Detection of COVID-19 Cases from Chest X-Ray Images

Linda Wang, Alexander Wong|arXiv (Cornell University)|Mar 22, 2020

COVID-19 diagnosis using AI48 references267 citations

TL;DR

COVID-Net is an open-source, tailored CNN for detecting COVID-19 from chest X-ray images, built via a human–machine design process and evaluated on COVIDx, with an explainability audit validating its decision basis.

ABSTRACT

The 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. 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.

Motivation & Objective

• Motivate rapid, accessible screening of COVID-19 using chest X-ray imaging in resource-constrained settings.
• Introduce COVID-Net, a tailored CNN designed for COVID-19 detection from CXR images.
• Create COVIDx, a large open-access CXR benchmark dataset for COVID-19 research.
• Enable explainability analysis to audit model decisions and improve clinical trust.

Proposed method

• Develop an initial human-designed residual prototype for tripartite prediction (normal, non-COVID19 infection, COVID-19).
• Use machine-driven design exploration (generative synthesis) with COVID-19 sensitivity ≥ 80% and PPV ≥ 80% as constraints.
• Construct COVID-Net with lightweight projection–expansion–projection (PEPX) blocks and selective long-range connectivity.
• Pretrain on ImageNet and train on COVIDx with Adam, learning rate 2e-4, 22 epochs, batch size 64, plus data augmentation and batch re-balancing.
• Audit model decisions with GSInquire explainability to ensure reliance on relevant lung regions rather than artifacts.

Experimental results

Research questions

• RQ1 Can COVID-Net achieve high sensitivity and PPV for COVID-19 detection from chest X-ray images on the COVIDx dataset?
• RQ2 Do architectural design choices from the machine-driven exploration (PEPX blocks, selective connectivity, and architectural diversity) improve performance and efficiency versus standard architectures?
• RQ3 How well does explainability (GSInquire) reveal the true factors behind COVID-Net predictions and validate trustworthy decisions?

Key findings

• COVID-Net achieves 93.3% test accuracy on COVIDx.
• COVID-Net attains 91.0% sensitivity for COVID-19 and 98.9% PPV for COVID-19.
• COVID-Net uses ~11.75 million parameters and ~7.50 G MACs, with far lower complexity than VGG-19 and ResNet-50.
• COVID-Net delivers higher COVID-19 sensitivity and PPV than VGG-19 and ResNet-50 in the comparisons.
• GSInquire explainability shows COVID-Net relies on lung regions for detection, supporting trustworthy reasoning.