In this study, we combined optical technology and machine learning to classify dental problems.We took totally 16 dental samples and 79 OCT images including 32 dental calculus(CA) images and 47 normal (HC) images. After image processing, we obtained optical attenuation coefficient, surface roughness and spectral information, and we put these features into two layer neural networks for training. We divided the data into training (24 CA / 37 HC = 61 total) and test (8 CA / 10 HC = 18 total) data, and the training data was checked with 10-fold cross validation to confirm no over-trained. The results showed that the model validity is 78%, and the test results have a sensitivity of 86%, specificity of 100%, and total accuracy of 94%.
|