An Advanced Deep Learning Approach for Primary Osteoporosis Prediction Using Radiographs with Clinical Covariates

• Published in: 2023 7th International Conference on Electronics, Communication and Aerospace Technology (ICECA) pp. 788 - 793
• Main Authors: Umamaheswari, R., Lakshmi, D., Pandi, V.Samuthira, Geetha, B., Sumithra, S., Y, Ragini P
• Format: Conference Proceeding
• Language: English
• Published: IEEE 22.11.2023
• Subjects: Automatic detection; Convolutional Neural Network; Convolutional neural networks; Deep learning; Dentistry; Diagnostic radiography; Osteoporosis; Osteoporosis prediction; Panoramic radiographs; Predictive models; Risk management
• DOI: 10.1109/ICECA58529.2023.10395285

A growing worry for world health is osteoporosis as a result of longer life expectancies. Its early detection, however, is still difficult because the signs are frequently lacking. As a result, using frequently used dental panoramic radiography for osteoporosis screening has shown to be both beneficial and cost-effective. This study examines the use of deep learning for osteoporosis categorization using panoramic dental radiographs. The study also investigates how the addition of clinical covariate data to radiographic image affects the accuracy of identification. Numerous clinical techniques have been created to evaluate the risk of osteoporosis and help choose postmenopausal women for measurements of bone mineral density. This study makes a contribution by developing and evaluating deep learning models to improve the precision of osteoporosis risk assessments using knee images. The application of Inception-v3 deep learning models allows for the evaluation of osteoporosis using knee panoramic radiographs. When compared to individual Convolutional Neural Networks (CNNs), the ensemble model exhibits appreciable improvements in performance across all metrics, especially in accuracy. The results highlight the accuracy of CNN-based deep learning in identifying osteoporosis from knee panoramic radiographs. The study also emphasizes the opportunity for accuracy enhancement using an ensemble deep learning Inception-v3 model that incorporates patient variables, which achieved 94.5% accuracy rate.