Swarm intelligence-driven HC2NN model for optimized COVID-19 detection using lung imaging

Abstract

COVID-19 virus has emerged as a formidable global health challenge, significantly complicated by the continuous evolution of viral variants that modify the virus’s structural characteristics. Predicting disease affection in the lungs is complex and degrades the accuracy level of COVID-19 variants through imaging techniques, which remains a formidable challenge. The complexities involved in identifying these regions are exacerbated by issues such as image degradation, the high dimensionality of features, and scaling properties, all contributing to an increased rate of false positives. Consequently, this leads to decreased disease detection frequency, lower precision accuracy, and poor performance of traditional diagnostic methods, as reflected in reduced F1 scores and overall detection accuracy. To resolve this problem, enhanced Swarm intelligence-based optimal feature engineering with Hyperscale capsule net-convolution neural network (HC2NN) was used to identify the survey of COVID-19 affection accurately. The preliminary process takes place to preprocess the covid-variant dataset with the support of adaptive Gaussian with wavelet filters. Then, Iterative Intra Subset Object Scaling (I2SOS) is applied to identify the disease-affected region. Then, interrogative slice fragment clustering (ISFC) is used to segment the disease region. Throughout the disease region properties, the feature selection is applied with Swarm intelligence, and identification is carried out by HC2NN work to effectively find the disease margin. The proposed experiment results project higher precision accuracy in prediction rate as well as in increasing true positives rate to attain the best recall, sensitivity performance, and F1 score. The novelty proves to have a higher performance than the existing traditional methods.