KyrgyzDentex: A Unique Multi-Modal Dental Radiograph Dataset from Kyrgyzstan with Classical Machine Learning and Deep Learning Benchmarks for Caries Detection
Year: 2026
Type: conference-paper
Abstract: Dental caries remain one of the most prevalent chronic diseases worldwide, yet timely diagnosis in low-resource settings such as Kyrgyzstan is hampered by limited access to trained dental professionals. We introduce KyrgyzDentex — the first publicly accessible, unique Kyrgyz dental radiograph dataset, comprising 3,848 periapical and 1,376 panoramic images, of which 564 periapical and 1,366 panoramic are annotated with bounding-box labels for carious lesions. We evaluate two complementary tasks: (1) binary classification and (2) object detection. For classification we compare four classical machine learning algorithms with two deep learning architectures (ResNet-50, EfficientNet-B0). For detection we benchmark YOLOv8n, YOLOv8s, and Faster R-CNN. EfficientNet-B0 achieves the best classification performance (F1= 0.658, AUC = 0.781 on periapical images). Object detection remains challenging, with YOLOv8s reaching [email protected] = 0.052 on panoramic images. The dataset and code are publicly available at https://huggingface.co/datasets/TTimur/dental-cavity-kg and https://github.com/golden-ratio/dental_cavity_kg.
Language: English
Subfield: Oral Surgery
Field: Dentistry
Domain: Health Sciences
Open Access status: hybrid
OpenAlex ID: W7212320620
