A multilevel neural network model for density volumes classification
Sergio Di Bona, Gabriele Pieri, Ovidio Salvetti · 2002
The accurate detection of tissue density variation in CT/MRI brain datasets can be useful for analysing and monitoring pathologies with slight differences. In fact, the objective knowledge of density distribution can be related to anatomical structures and therefore the process of monitoring illness and its treatment can be improved. In this paper, we present an approach for the classification of tissue density in three dimensional brain tomographic scans. The proposed approach is based on a hierarchical neural network model able to classify the single voxels of the examined datasets. The approach has been evaluated on both normal and pathological cases selected by an expert neuroradiologist as study cases. The results have shown that the method has a good effectiveness in practical applications and that it can be used for designing a full 3D instrument suitable for supporting the analysis of disease diagnosis and follow-up.