Models of neural networks for image analysis of cintillography from the urinary system

From the several available diagnostic techniques, the non-invasive ones are the less aggressive to the human body. The use of radionuclides for renal function evaluation seems to be more secure than the other diagnostic procedures. However, the use of radioactive materials, makes it necessary to maximize the beneficial effects of this technique, conducing inevitably to the development of support tools for the analysis of their results. The objective of this work is to improve a system model for image analysis of scintillation. The detection and identification of renal anomalies is done by means of analysis of parts of the image, taking in account the spatial distribution of those anomalies. Geometric operations in conjunction with the segmentation technique enable the improvement of image analysis information processing. In this work, we look for the best artificial neural network architectures to introduce the knowledge of the medical expert into the diagnostic system. The search of the best neural network architectures, the geometric techniques and the image segmentation technique permitted a significant improvement of diagnostic analysis model by using scintillation. (author)