부산대학교 도서관

The paper considers the creation of a system for determining pathology on chest radiographs based on convolutional neural networks and image processing using fuzzy sets. The topic of the work is relevant, since today methods based on convolutional neural networks are widely used to recognize objects in an image. The architecture of the system is based on several neural networks. At the first stage, image segmentation is performed to search for areas of interest, then classification and detection of found pathologies. The developed system makes it possible to determine pneumothorax, focal shadowing, pneumonia, atelectasis and hydrothorax. In addition, the system provides information about the exact localization of pathology. In the course of the work, tests were carried out on real X-ray images, for all classes, sensitivity and specificity of more than 0.9 were obtained.