A Novel Hybrid Approach for Multi-class Classification of Lung Related Diseases

Abstract Of late, the whole world has seen an alarming impact due to the widespread and rapid rise of Covid-19 cases. This paper addresses the challenge of early and accurate detection of lung-related diseases, par- ticularly Covid-19, Pneumonia, Lung-Cancer, and Normal chest images, using hybrid deep learning models. The study evaluates the perfor- mance of three hybrid deep learning architectures that use three feature extraction models, namely InceptionV3, ReNet50, and Xception net Con- volution Neural Network (CNN). The proposed hybrid models employ recurrent neural networks, namely CNN-LSTM, CNN-GRU, and CNN- BiGRU, for classification. The study uses both computed tomography (CT) and chest X-ray (CXR) images to evaluate the models’ per- formance. The CNN-BiGRU model achieved the highest accuracy of 99.57%, with an average sensitivity and precision of 99.00%, and an F1-score of 99.00%. The study finds that the hybrid deep learning strategy produces highly competitive accuracy while requiring modest training time and loss for the classification of lung-related diseases.