Deep Learning Based Image Segmentation for Detection of Odontogenic Maxillary Sinusitis

Abstract

The aim of this study is to develop a new approach for Computed Tomography (CT) image segmentation based on Convolutional Neural Network (CNN) for detection of Odontogenic Maxillary Sinusitis (OMS). Our study is based on results of examination of 100 people (320 CT). Using the RadiAnt software, we selected tomographic scans on which the location of the roots of the teeth in the maxillary sinus was clearly visualized. Next step has been split into tasting and training sets (30% and 70%). An image preprocessing techniques have been applied. The architecture of the convolutional neural network U-NET was used for the image segmentation. On its basis 6 models with different values of hyperparameters (such as Batch Size, Epochs, Validation Split) were built and a comparative analysis of the results of the training was done.