Instance Segmentation and Labeling of Teeth from Dental X-Ray using Region Based Convolutional Neural Network

Abstract

Radiological Examination of teeth is a primary step that a dentist usually takes to diagnose the problem before further treatment. The diagnosis involves searching for diseases ranging from cavities to tumors, So, correct diagnosis is vital for timely and precise treatment. This paper attempts to solve one of the elementary steps in diagnosis i,e, Labeling of Teeth, using Region-Based Convolutional Neural Networks that help reduce monotonous work for a dentist and provide segments of each tooth for further diagnosis of diseases with the use of Mask R-CNN. We used 200 panoramic X-Ray images of 4 categories to train, test and validate the model. Mask R-CNN with pre-trained weights of COCO Dataset is employed. We further tuned the weights of the dental X-ray dataset considered in the paper for better performance. On testing the learned model, the performance measures were encouraging.