Akademik Veri Yönetim Sistemi
6th International Congress on Human-Computer Interaction, Optimization and Robotic Applications, HORA 2024, İstanbul, Türkiye, 23 - 25 Mayıs 2024
Early diagnosis and accurate classification of diseases significantly impact treatment processes. The proposed study focuses on the regional detection of ectopic eruption of first permanent molars and severity classification based on their condition, using the deep learning method YOLOv8. The study utilized panoramic X-ray images from 168 patients aged 5-9 years. Images were labeled by expert dentists according to the ectopic eruption conditions (mild-normal-moderate-severe). Four scenarios were created in the study. The model trained using data augmentation methods, including the classification of ectopic eruption severity, showed the lowest performance with an accuracy rate of 55.76%. The model trained on only the original dataset, also including severity classification, showed an accuracy rate of 62.54%. In scenarios where severity classification was not considered, and the dataset was organized as Ectopic Eruption present/absent, the version including the mild class achieved an accuracy rate of 72.75%, while the approach excluding the mild class achieved an accuracy rate of 71.76%. The findings indicate that the YOLOv8 model can be an effective tool for detecting and classifying ectopic eruption conditions of first permanent molars.