Optimizing Deep Neural Networks with Differential Evolution for COVID-19 Diagnosis

Abstract

Accurate detection of COVID-19 remains a
significant challenge, particularly due to the inherent
limitations in conventional diagnostic methods such
as RT-PCR testing. This study introduces two
deep learning-based classification strategies using the
COVIDGR chest X-ray data set. The first approach
leverages the ResNeXt50 architecture, enhanced through
training hyperparameter optimization via the Differential
Evolution algorithm. The second strategy also employs
DE to design custom CNN architectures based on the
Multi-scale Feature Learning model, optimizing structural
hyperparameters. Experimental results show that both
methods outperform their non-optimized counterparts as
well as several state-of-the-art approaches, achieving an
overall performance of 90.32%. These findings highlight
the potential of DE as a powerful tool for improving
automated diagnosis of respiratory diseases.