Abstract
Computer Vision, as an area of Artificial Intelligence, has recently achieved success in tackling numerous difficult challenges in health care and has the potential to contribute to the fight against several lung diseases, including COVID-19. In fact, a chest X-ray is one of the most frequent radiological procedures used to diagnose a variety of lung illnesses. Therefore, deep learning researchers have recommended that deep learning techniques can be used to build computer-aided diagnostic systems. According to the literature, there are a variety of CNN structures. Unfortunately, there are no guidelines for compressing these architectural designs for any particular task. For these reasons, this design is still very subjective and hugely dependent on data scientists' expertise. Deep convolution neural networks have recently proven their capacity to perform well in classification and dimension reduction tasks. However, the problem of selecting hyper-parameters is essential for these networks. This is due to the fact that the size of the search space rises exponentially with the number of layers, and the large number of parameters requires extensive calculations and storage, which makes it unsuitable for application in low-capacity devices. In this paper, we present a system based on a genetic method for compressing CNNs to classify radiographic images and detect the possible thoracic anomalies and infections, including the case of COVID-19. This system uses pruning, quantization, and compression approaches to minimize the network complexity of various CNNs while maintaining good accuracy. The suggested technique combines the use of genetic algorithms (GAs) to execute convolutional layer pruning selection criteria. Our suggested system is validated by a series of comparison experiments and tests with regard to relevant state-of-the-art architectures used for thoracic X-ray image classification.