Oliveira, Danilo, Barreto, J. B., Mesquita, I. M., Paula Jr, I. C., Chaves, F. N., Sampieri, M. B. S., and Madeiro, J. P.
ABSTRACTThe third molar tooth (or wisdom tooth) is considered potentially problematic and its extraction is always in debate among dentists due to the fact that pathologies can arise with its permanence. In order to aid diagnosis, the present work seeks to automate the detection of cysts in these teeth using radiographic images. For this, four Convolutional Neural Networks (CNN) architectures are analysed for classification, two classical architectures and two proposed ones with experiments about image pre-processing techniques. The best results were reached with one of the proposed architectures, which use the growth of filter numbers along the convolution layers, using gamma correction pre-processing. They obtained better performance, with an accuracy of 81.99%, a precision of 78.90%, a sensitivity of 86.17% and an F1-score of 82.19%. The results identified the best combination of architecture and pre-processing and demonstrate that the proposal allows for automation in the diagnosis of cysts.