Пошук каріозних захворювань на зображеннях за допомогою штучного інтелекту

Abstract

This work is aimed at using artificial intelligence to detect carious diseases in the human oral cavity using photographs. The ICDASTM system is used to classify caries, which has four classes of caries, namely: sound surfaces, Initial stage caries, Moderate stage caries and Extensive stage caries. The paper compares the effectiveness of the following algorithms: RetinaNet, YOLOv3, SSD and Faster R-CNN. The comparison is made according to the following indicators: sensitivity (TPR), specificity (TNR), accuracy (Accuracy) and accuracy of prediction (Precision). The Faster R-CNN model was the most accurate, while the SSD model was the least accurate