Electronic Theses and Dissertation

Abstrak – karies gigi merupakan salah satu penyakit gigi yang paling umum dan sering terlambat terdeteksi dikarenakan kualitas citra yang rendah, seperti kontras yang kurang baik, noise, serta batas lesi yang samar. meskipun berbagai metode deep learning telah dikembangkan untuk mendeteksi karies, performanya masih dipengaruhi oleh kualitas citra masukan sehingga proses ekstraksi fitur belum optimal. penelitian ini bertujuan meningkatkan kualitas citra klinis melalui integrasi histogram equalization (he) dan canny edge detection sebelum proses deteksi objek menggunakan arsitektur rf-detr. he digunakan untuk meningkatkan kontras citra, sedangkan canny edge detection diterapkan untuk memperjelas informasi tepi lesi karies dan mengurangi noise. kinerja metode dievaluasi menggunakan metrik evaluasi kualitas citra serta performa deteksi objek berbasis mean average precision (map). hasil penelitian menunjukkan bahwa model dengan integrasi he dan canny edge detection memperoleh nilai map50 sebesar 0,911, lebih tinggi dibandingkan dataset tanpa praproses (0.872), dataset dengan he (0.878), maupun dataset dengan canny edge detection (0.903). hasil tersebut menunjukkan bahwa integrasi kedua teknik praproses mampu meningkatkan kualitas informasi visual citra sehingga mendukung peningkatan performa deteksi objek berbasis rf-detr pada citra klinis karies gigi. kata kunci : karies gigi, deteksi karies gigi, image enhancement, edge detection, rf-detr.

Abstract – Dental caries is one of the most common dental diseases and frequently detected late due to low image quality, such as poor contrast, noise, and faint lesion boundaries. Although various deep learning methods have been developed to detect caries, their performance is still affected by the quality of the input images, leading to suboptimal feature extraction. This study aims to improve the quality of clinical images through the integration of Histogram Equalization (HE) and Canny Edge Detection before the object detection process using the RF-DETR architecture. HE is applied to enhance image contrast, while Canny Edge Detection is applied to sharpen the edge information of caries lesions and reduce noise. The performance of the method is evaluated using image quality evaluation metrics alongside object detection performance based on mean Average Precision (mAP). The results show that the model integrated with HE and Canny Edge Detection achieved mAP value of 0.911, which is higher than the dataset without preprocessing (0.872), the dataset with HE alone (0.878), and the dataset with Canny Edge Detection alone (0.903). These results show that the integration of both preprocessing techniques successfully enhances the quality of visual image information, thereby supporting the performance improvement of RF-DETR object detection model in clinical images of dental caries. Keywords : Dental caries, dental caries detection, image enhancement, edge detection, RF-DETR.