Osteoporosis Detection from Dental Periapical Radiographs Using Deep Learning Approaches

Osteoporosis is a systemic skeletal disease marked by low bone mineral density and microstructural deterioration, leading to increased fracture risk in the elderly. Standard diagnosis via dual-energy X-ray absorptiometry (DXA) is costly and not widely available, motivating alternative screening methods. Dental radiographs have emerged as a promising source of osteoporosis indicators especially in individuals with oral cancer, since changes in mandibular trabecular bone and cortical width on radiographs correlate with skeletal bone density. In this work, a custom Convolutional Neural Network (CNN) and a ResNet50 model are proposed to classify cropped 80 × 80 regions of interest (ROIs) from dental periapical X-rays into three categories: normal, osteopenia, and osteoporosis. We use the publicly available Dataset of Dental Periapical Radiograph for Osteoporosis Classification (postmenopausal Javanese women, age >50) with DXA-derived labels. Our custom CNN achieves a higher classification performance with F1-scores of 0.92 for normal, 0.97 for osteopenia, and 0.99 for osteoporosis. These results represent a substantial improvement over recent state-of-the-art benchmarks and demonstrate the feasibility of deep learning on periapical radiographs for osteoporosis screening. The dataset we use is distributed under a CC BY 4.0 license.