Line Operator as Preprocessing Method for CNN-based Osteoporosis Detection in Dental Panoramic Radiograph

Abstract

Osteoporosis is a disease that can be detected via the trabecular bone pattern in Dental Panoramic Radiograph (DPR). Trabecular bone pattern is difficult to see by the naked eye due to the low contrast and low resolution of DPR. This can affect the performance of osteoporosis disease detection using Convolutional Neural Network (CNN). In this paper we propose the use of Line Operator (LO) on DPR images as a preprocessing method to enhance trabecular bone pattern for CNN-based osteoporosis detection. LO is a method that can enhance line-like structures in medical images such as retina and DPR dataset. To study the effect of LO on CNN-based osteoporosis detection, the performance of non-preprocessed images, LO-preprocessed images and LO + histogram equalization pre-processed images was compared. Results showed that LO-preprocessed images give best osteoporosis detection accuracy of 0.875