Diagnostika idiopatické skoliózy

Abstract

The thesis deals with the diagnosis of idiopathic scoliosis. The main aim of the work was to evaluate the diagnosis of idiopathic scoliosis from retrospective patient records. The thesis is divided into theoretical and practical parts. The theoretical part describes the anatomy of the spine, idiopathic scoliosis and its symptoms, etiology, diagnostic methods and treatment. The theoretical part also includes a basic description and application of artificial intelligence, including a detailed description of the neural network, which was modelled within the practical part of the thesis. In the practical part, the aim was to design and implement a statistical evaluation of the success of diagnostics and the degree of idiopathic scoliosis. This evaluation was done on patient records selected by an expert and these records were also used to design neural networks. Two models of neural networks have been created to diagnose idiopathic scoliosis, namely the forward neural network and the PatternNet neural network. Neural networks have one hidden layer that contains 10, 30, 50, 100 and 150 neurons. The entry for diagnosis was 12 risk factors together with knowledge of the resulting diagnosis. The outcome is probability values that predict diagnosis into three classification classes. The quality criterion for the classification of idiopathic scoliosis in the feed forward neural network was the root mean square error, and in the case of the PatternNet neural network, the cross entropy. Neural networks were modeled using the MATLAB 2021a system.