Deep neural network model with Bayesian optimization for tuberculosis detection from X-Ray images

Tuberculosis is a chronic lung disease caused by bacterial infection, and more than 10 million people get this disease every year, especially in developing countries. Early diagnosis of tuberculosis is important for effective treatment. Thus, a new approach for diagnosing tuberculosis disease is proposed in this paper, which is based on the development of a deep neural network (DNN) model in which hyperparameters are determined using the Bayes optimization method. First, feature extraction was conducted using pre-trained deep learning models such as VGG16, EfficientNetB0, ResNet101, and DenseNet201 architectures in the proposed approach. Following that, four DNN models in which hyperparameters were selected using the Bayesian optimization method were developed utilizing these features extracted from pre-trained deep learning architectures. Finally, these DNN models were used to classify tuberculosis disease, and the classification performance of the developed models was compared. The results showed that the EfficientNetB0 model yields the best performance with 99.2857% accuracy, followed by VGG16 with an accuracy of 97.9286% and DenseNet201 with an accuracy of 97%. The ResNet101 model has the lowest accuracy with an accuracy of 95.6429%. Consequently, the best pre-trained model for extracting features from images as well as the most efficient and effective DNN structure for detecting tuberculosis disease has been revealed in this study.