A master's thesis in the Department of Computer Science at the University of Technology classified lung diseases using wavelet scattering and a twisted neural network.
The thesis submitted by the student, Sufian Othman Zabin, aimed to classify chest x-ray images of lung diseases to help physicians and radiologists in diagnosis. You need automated and intelligent systems that accurately classify chest x-rays as normal, pneumonia, and COVID-19 used powerful, deep networks in classification tasks called the wavelet scattering network and the convolution neural network, effectively classifying chest X-ray images for use in the medical field.
The letter showed a classification of lung diseases using a wavelet scattering network (WSN) and a convolution neural network (CNN) based on medical X-ray images. The proposed system, which uses the Cohen-Kaggle dataset, aims to classify lung diseases into normal diseases, pneumonia, and COVID-19. The system also uses data augmentation and preprocessing, including resizing, to prepare an image for feature extraction using either a scattering network. Wavelet, convolution neural network, or a combination of WSN and CNN. Soft Max classification was used to classify lung disease on input x-ray images, into normal and abnormal conditions, pneumonia, and COVID-19.
The three proposed methods systems used X-ray images of the data set to perform feature extraction, and the accuracy of each method is calculated and discussed during this study. The system that used wavelet scattering network for feature extraction achieved 97% accuracy, while the system using CNN achieved 96%. However, the system that used a combination of the two methods was the most accurate, achieving an accuracy of 98%.
The researcher reached the following conclusions: The combination of the wavelet scattering network and the convolution neural network are two classifications of lung diseases by means of X-ray images to obtain high accuracy. Features extracted from CXR images were also managed using wavelet scattering and CNN. The results showed that the wave scattering network was more accurate (97%) than the CNN network (96%), although the training took longer. However, the combination of the two systems achieved an accuracy of 98%, which was better than both single methods in terms of accuracy and time.
The discussion committee consisted of Prof.Dr. Yousry Hussein Ali as chairperson, Prof. Dr.Rahim Abdel-Saheb, and Prof. Dr. Mustafa Salam, as members of Prof. Dr. Iqbas Ezz El-Din.
