City imaging specialist evaluates the performance of five deep learning architectures in classifying COVID-19, Pneumonia and Healthy Individuals

In his recent research paper (co-authored with MSc in AI student Azucena Ascencio-Cabral), Dr Constantino Carlos Reyes-Aldasoro (medical imaging and data visualisation specialist in City's giCentre), evaluates the performance of five deep learning architectures in classifying COVID-19, Pneumonia and Healthy Individuals with computed tomography and proposed a framework for a fair comparison of these architectures.

One of the classifiers he used is the Residual Network (ResNet) - a deep learning model which is used for computer vision applications.

Front cover of journal

Other classifiers used were ResNet-50r, DenseNet-121, MobileNet-v3 and the state of the art CaiT-24-XXS-224 (CaiT) transformer.

The research paper, Comparison of Convolutional Neural Networks and Transformers for the Classification of Images of COVID-19, Pneumonia and Healthy Individuals as Observed with Computed Tomography, was published in, and selected as the front cover for the Journal of Imaging.

Computed tomography (CT) is a medical imaging procedure using special x-ray equipment to acquire detailed three-dimensional scans of organs inside the body.