COVID-19 classification

Customer: Beth Israel Deaconess Medical Center (Boston, United States)

Summary: The novel coronavirus 19 (COVID-19) continues to have a devastating impact around the globe, leading many scientists and clinicians to actively seek new techniques to help combat this disease. Modern machine learning methods have shown promise in their application to support the healthcare industry through data and analytics-driven decision-making, inspiring researchers to develop new angles to fight the virus. In this study, we aim to develop a CNN-based method for the detection of COVID-19 using patients' chest X-rays. Building on the incorporation of convolutional units, the proposed method makes use of indirect supervision based on Grad-CAM. This technique is used in the training process, where Grad-CAM's attention heatmaps support the network's predictions. Despite recent progress, the scarcity of data has limited the development of a robust solution. We extend existing work by combining publicly available data from five different sources and carefully annotating the resulting images into three categories: normal, pneumonia, and COVID-19. To achieve high classification accuracy, we propose a training pipeline based on indirect supervision of traditional classification networks, where guidance is provided by an external algorithm. With this algorithm, we observed that widely used standard networks can achieve an accuracy comparable to tailor-made models, specifically for COVID-19. One network in particular, VGG-16, outperforms the best of the tailor-made models.

Collaborators: Oleg Talalov (Amazon, Vancouver, Canada), Alexander Proutski (Quantori, Hague, Netherlands), Diana Litmanovich (Beth Israel Deaconess Medical Center, Boston, United States), Alexander Kirpich (Georgia State University, Atlanta, United States), Yuriy Gankin (Quantori, Cambridge, United States)

Project type: Commercial / Research

Media: Journal paper, Preprint

Figure 1. Proposed pipeline based on guided attention