Detection and classification of pneumonia using novel Superior Exponential (SupEx) activation function in convolutional neural networks

Many diseases threaten human health in many ways. Pneumonia is accepted as one of the most important health problems and causes of death worldwide. Therefore, deep learning techniques are used to solve problems in various fields, including diagnosing pneumonia. Convolutional neural networks (CNN), an artificial intelligence technique, are widely used in many areas such as segmentation, classification, signal processing. In deep networks, activation functions have an import. In CNN architectures, while the activation functions (AFs) are being developed, the activation functions are developed by taking into account the features such as not getting stuck in the local minimum of the CNN model and increasing the training performance. In the literature, although the ReLU is used in most studies, ReLU faces the problem that negative weights cannot be added to the network. To overcome the problem, the effect of the proposed new AFs on the real-world problem was investigated. In this study, a method for detecting pneumonia is suggested based on the novel proposed activation function and different CNN models. In the experimental studies, pneumonia was detected from chest X-ray images using ReLU, LReLU, Mish, Sigmoid, Swish, Smish, Logish, Softplus and the proposed activation function Superior Exponential (SupEx) were compared. Additionally, experimental study was performed through the MNIST and CIFAR-10 datasets to support the success of the proposed SupEx AF on the results. In the results obtained, it is seen that the CNN models used with the proposed SupEx have the best performance in both the detection of pneumonia from chest X-ray dataset and the MNIST dataset. The outstanding aspect of this study is that it is proposed to detect pneumonia with a new activation function, and it has been demonstrated that this activation function also works successfully in traditional benchmark datasets.