Abstract
Today, the aid of modern technology in the medical field can be seen in every respect. In the field of radiology, many deep learning and image processing techniques have been applied for timely and better analysis, as well as conclusive results. However, dental radiographs are detailed, and some of these details are fine and vague, making them difficult to interpret. With the help of deep learning techniques, the automated uncovering of these fine details has great potential. In this paper, we aim to detect pulp stones in dental radiographs using Convolutional Neural Network (CNN)-based feature extraction followed by multiple classifiers. We conclude that the Residual Network 50 (ResNet-50) achieves an accuracy of 76.4% with the Medium Gaussian Support Vector Machine (SVM), while Inception v3 reaches an accuracy of 73.1% with the same classifier. Further, the ResNet-50's false positive rate is less than the Inception v3's by 7%, giving it the potential for further experiments.