Deep learning models comparison for tissue classification using optical coherence tomography images: toward smart laser osteotomy

We compared deep learning models as a basis for OCT image-based feedback system for smart laser osteotomy. A total of 10,000 OCT image patches were acquired ex-vivo from pig's bone, bone marrow, fat, muscle, and skin tissues. We trained neural network models using three different input features (the texture, intensity profile, and attenuation map). The comparison shows that the DenseNet161 model with combined input has the highest average accuracy of 94.85% and F1-score of 94.67%. Furthermore, the results show that our method improved the accuracy of the models and the feasibility of identifying tissue types from OCT images. (C) 2021 Optical Society of America under the terms of the OSA Open Access Publishing Agreement