Canis collie optimization-based federated deep neural network for classification of pulmonary infections through chest CT scans

Contagious lung infections are a matter of serious concern as they spread rapidly from person to person through aerosols and small droplets. Identifying the type of pulmonary infection is crucial due to similar clinical manifestations and an overlap of CT imaging characteristics of infections such as COVID-19, pneumonia and tuberculosis. Clinicians have relied on deep learning-based automated systems as a supplementary diagnostics tool. In recent times, federated learning has been extensively used in COVID-19 diagnosis to protect privacy of patient data sourced from multiple medical facilities. The primary intent of this research is to examine the efficacy of federated learning in a multi-class environment. This study developed a federated deep convolutional neural network to distinguish between COVID-19, common pneumonia and normal controls. Robust texture feature extraction, federated learning and an innovative hybrid bio-inspired optimizer are the salient features of our proposed model. The proposed Canis collie optimizer, a fusion of grey wolf optimizer and the border collie optimizer is used to fine-tune the parameters of the 19-layered federated deep convolutional neural network. When tested on an open access dataset, our model achieved maximum accuracy, recall and specificity of 99.49%, 99.46%, and 99.60%, respectively. Our proposed system outperformed the other cutting-edge research works that have implemented federated learning for detection of lung infections.