Identificação de regiões císticas dentárias em imagens de radiografia panorâmica utilizando algoritmos de segmentação e classificação por textura

Abstract

The dental cyst is a pathology that presents characteristics of expansion and bone absorption, its early identification is fundamental for the treatment. Several computer vision systems seek to carry out the identification of cysts. Therefore, the present research aims to provide a methodology for automatic identification of cystic regions in radiographic examinations. For this, investigations of initial segmentation algorithms and texture descriptors applied in a neural network are carried out. The best initial segmentation results were from the Otsu multilevel, with accuracy of 68.56%, sensitivity of 66.29% and specificity of 68.10%. The best mean final classification results were LBP, with 92.97% accuracy, 89.24% sensitivity and 96.68% specificity.