Two-dimensional automatic SSA domain multi-modal deep neural network for detection of COVID-19 from lung ultrasound images

Abstract

This chapter proposes an image decomposition-based multi-modal deep convolutional neural network (CNN) for the automated detection of COVID-19 using ultrasound images. The two-dimensional (2D) automatic-singular spectral analysis (Auto-SSA) is introduced to decompose ultrasound images into four modes or sub-images. The obtained modes are then used as input to the proposed multi-modal CNN model for COVID-19 detection. The performance of the proposed model is assessed on a dataset consisting of 3710 ultrasound images. The classification schemes such as COVID-19 versus pneumonia versus other ailments and COVID-19 versus pneumonia versus healthy are considered in this work. The proposed multi-modal deep CNN has obtained the maximum accuracy values of 100% and 99.87% for COVID-19 versus pneumonia versus other ailments-based classification schemes using 5-fold cross-validation (CV) and hold-out validation techniques. © 2024, IGI Global. All rights reserved.