AI-based diagnosis of COVID-19 patients using X-ray scans with stochastic ensemble of CNNs

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Deep Learning

Background:

• The COVID-19 pandemic presents significant global health and economic challenges.
• Rapid and accurate diagnosis is crucial for managing the overwhelming number of cases and easing pressure on medical facilities.
• Existing diagnostic methods require enhancement to meet the demands of large-scale outbreaks.

Purpose of the Study:

• To develop a rapid and scalable diagnostic system for COVID-19 using medical imaging.
• To improve the discriminability of features in medical images through a novel deep learning approach.
• To evaluate the model's effectiveness on diverse datasets including chest X-rays and CT scans.

Main Methods:

• A stochastic deep learning model was proposed, constraining deep representations over a Gaussian prior.
• The model learns a latent space from X-ray image distributions using an ensemble of convolutional neural networks.
• Predictions are generated by regressing outputs from an ensemble of classifiers utilizing the latent vector.

Main Results:

• The model achieved an overall accuracy of 0.91 and an Area Under the Curve (AUC) of 0.97 for classifying COVID-19, normal, and pneumonia from X-rays.
• Experiments on a large chest X-ray dataset demonstrated robust classification for Atelectasis, Effusion, Infiltration, Nodule, and Pneumonia.
• The proposed model exhibited a superior understanding of X-ray images, indicating its potential for broader medical image analysis applications.

Conclusions:

• The developed stochastic deep learning model provides a fast and scalable solution for COVID-19 diagnosis.
• The model's ability to enhance feature discriminability makes it effective for medical image analysis.
• Its generic nature suggests applicability to various domains within medical imaging beyond COVID-19 detection.