Deep-Learning Segmentation and Recognition of Tooth in Thresholded Panoramic X-ray

Timely detection and handling of dental issues are essential, and clinical-level recognition of dental issues depends on a personal check by the dentist and imaging-supported screening. Panoramic X-rays are widely adopted in clinics to detect the various abnormalities associated with teeth, and efficient examination is necessary to plan appropriate treatment. The proposed research aims to develop a deep-learning scheme to automatically examine the tooth and its condition using the Panoramic X-ray Image (PXI). The various phases involved in the proposed methodology include; (i) Image collection and resizing, (ii) Otsu's thresholding with Butterfly-Algorithm to enhance the tooth regions in PXI, (iii) VGG-UNet segmentation of tooth sections, and (iv) YOLO-V3 based recognition of individual tooth in the chosen test images. This work considered the 1000 images of the UFBA-UESC PXI dataset. It implemented a series of operations, such as image resizing, tooth region enhancement with thresholding, deep-learning-based segmentation, and tooth detection. The experimental outcome was improved when a VGG-UNet was implemented (accuracy = 98.84%). Further, the proposed YOLO-V3 efficiently detects the individual tooth from the segmented images.