Classificação de imagens utilizando redes de Kolmogorov-Arnold para diagnóstico de doenças pulmonares

Abstract

This work aims to develop a classification system for images obtained from chest x-ray for the diagnosis of lung diseases Pneumonia and Tuberculosis, using algorithms based on the Kolmogorov-Arnold Network (Efficient and Fast). The research begins with a theoretical review of lung and respiratory system diseases, establishing the necessary foundation for the application of machine learning techniques. The methodology involves the pre-processing of images through data augmentation techniques, the construction of customized models based on CNN, Efficient KAN and Fast KAN. The performance of the models is evaluated by evaluation metrics, allowing the comparison of the results obtained. These results indicate that the architectures based on the Kolmogorov-Arnold Network proved to be promising compared to the tests carried out for the CNN model, where it was shown that the Fast KAN technique, with a shorter time, equaled the percentages of the evaluation metrics with the CNN algorithm, with a difference of 12 hours.