Leveraging hybrid deep learning approaches for accurate cxr classification

OBJECTIVE The similarities between bacterial and COVID-19 pneumonia patterns in chest X-ray images (CXR) pose a challenge in enhancing diagnostic rates. This study aims to apply a hybrid deep learning method combined with a machine learning classifier to accurately classify normal, bacterial, and COVID-19 pneumonia CXR. MATERIALS & METHODS An open public dataset was utilized to investigate hybrid AI models, consisting of 5260 CXR images. Sample sizes for normal, COVID-19, and bacterial pneumonia were 1792, 1658, and 1800 images, respectively. The hyperparameters included batch size, epoch size, and learning rates (0.001). The dataset was divided into 70%, 20%, and 10% for training, validation, and testing of the transfer CNN models, respectively. A hybrid AI model was designed to extract combined features from multiple CNN methods, and then these features were used to build a machine learning classifier. The classification performance was evaluated using accuracy and Kappa value. RESULTS The hybrid AI model provided an accuracy of 0.995 and a Kappa value of 0.991. Moreover, the precision among normal, COVID-19, and bacterial pneumonia groups were 0.993, 0.997, and 0.991, respectively. CONCLUSION In this study, the hybrid AI model combined with a machine learning classifier demonstrated reliable and accurate performance in classifying CXR images. For clinical application, the consideration of a larger CXR dataset is recommended for future research. The similarities between bacterial and COVID-19 pneumonia patterns in chest X-ray images (CXR) pose a challenge in enhancing diagnostic rates. This study aims to apply a hybrid deep learning method combined with a machine learning classifier to accurately classify normal, bacterial, and COVID-19 pneumonia CXR. An open public dataset was utilized to investigate hybrid AI models, consisting of 5260 CXR images. Sample sizes for normal, COVID-19, and bacterial pneumonia were 1792, 1658, and 1800 images, respectively. The hyperparameters included batch size, epoch size, and learning rates (0.001). The dataset was divided into 70%, 20%, and 10% for training, validation, and testing of the transfer CNN models, respectively. A hybrid AI model was designed to extract combined features from multiple CNN methods, and then these features were used to build a machine learning classifier. The classification performance was evaluated using accuracy and Kappa value. The hybrid AI model provided an accuracy of 0.995 and a Kappa value of 0.991. Moreover, the precision among normal, COVID-19, and bacterial pneumonia groups were 0.993, 0.997, and 0.991, respectively. In this study, the hybrid AI model combined with a machine learning classifier demonstrated reliable and accurate performance in classifying CXR images. For clinical application, the consideration of a larger CXR dataset is recommended for future research.