Neural networks-based tool for diagnosis of paranasal sinuses conditions

This paper describes the development of a neural networks-based software system for the analysis and diagnosis of sinus conditions. Traditional image processing techniques and artificial neural networks tools and algorithms such as the self-organizing maps (SOM) were invoked in the development of the diagnosis system. The data were in the form of anonymous CT-images of sinuses obtained from a local hospital. A major problem faced the development of the system was caused by the fragmented or incomplete boundaries between different objects in the CT-images. A new algorithm was developed and successfully applied to complete such boundaries. The system was thoroughly tested with real images and the results indicate a potential for the system to be integrated within a CT-scanning system to automate the process of diagnosis.