Abstract
Machine learning and augmented reality form very important computational tools in biomedicine, neurology and stomatology as well. The present paper is devoted to a novel method of spectroscopic detection of caries lesions that changes the optical properties of the affected tissue.
This method of the diffuse reflectance spectroscopy is used in many biomedical areas even though the analysis of associated data suffers from a large variance of acquired signals' shape and their properties. The proposed methodology of measured spectra analysis is based upon general methods of signal feature evaluation and the use of computational intelligence for their classification.
The paper compares properties of dental feature clusters for the set of 578 tissues with different levels of their changes. Classification results of selected features by the support vector machine, Bayesian method, k-nearest neighbour method and neural network enable to distinguish the healthy tissue and caries lesions with the accuracy from 94.1 to 98.4% and the cross-validation error lower than 8.3%.
These results suggest how the augmented reality and general mathematical signal processing methods can be beneficial for diagnostic purposes in dental research and possibly in the clinical practice as well.