This dissertation presents applications of Artificial Neural Networks to the image diagnosis process. Some concepts of Neural Networks are presented with foundations, learning algorithms and topologies used. Due to the need of pre-processing the input data, subjects related to the analysis of medical images from an expert point of view are considered. Among the analyzed aspects are the density, the form, the size and the location of these characteristic aspects of the image. Slight focus is given to the concept of space as well as to the process of digitalizing images adopted here. It was also prepared a routine that has the objective to facilitate handling and preparation of the images for the use with neural networks techniques. Synthetic data was elaborated and numeric experiments are presented using these data. Numeric experiments are presented with real data extracted from neurological images of Computerized Tomography, from images of the saturation of fluids in a porous environment generated in a CT Scanner, adapted for these analyses, property of CENPES - Petrobrás and also adontological images of simple x-rays showing abscesses, cists and granulated tissues. Both the tests with synthetic data and the tests with real data showed satisfactory results and, finally, some considerations are made on the perspectives of evolution of the method.