Quantum-Inspired Neural Networks for COVID-19 Detection Using Enhanced Vision Transformers

Abstract

The pandemic continues to present a serious challenge for the healthcare systems around the world and rapid and reliable diagnostics are crucial to aid clinical decision making. In this work, we present a novel framework called the Quantum-Inspired Vision Transformer (Q-ViT) for automated detection of COVID-19 in chest X-ray images. The proposed approach consists of an Enhanced Vision Transformer combined with a Quantum-Inspired Neural Network and a Quantum-Inspired Dropout mechanism to enhance the ability of the model to learn global features, model features interaction and resistance to classification errors. Experiments were carried out on the COVIDx CXR-3 data set which contains 30,386 chest X-ray images, among which 15,994 are COVID-19, 5,555 are Pneumonia and 8,837 are Normal. The proposed model outperformed the conventional optimization strategies such as SGDM and RMSprop with an accuracy of 89.00%, macro precision of 90.09%, macro recall of 89.00% and macro F1 score of 88.98%. The accuracies for the COVID-19, Pneumonia and Normal categories were 99.2%, 98.7%, and 99.0% respectively. The framework also had a sensitivity of 98.9% and specificity of 99.2%, demonstrating its potential for clinical application to computer-aided screening. While encouraging results, additional validation on external multi-center datasets is needed to ensure generalizability of results in the real-world and readiness for deployment.