Ensemble Deep learning model using panoramic radiographs and clinical variables for osteoporosis disease detection

Abstract

Worldwide, a large number of people suffer from the bone disease osteoporosis. Accurate diagnosis and classification are essential for managing and preventing many disorders. In order to classify bone density images into two categories—normal and osteoporotic—this study suggests a hybrid model that combines a multiclass Support Vector Machine (MSVM) with a Deep Convolutional Neural Network (DCNN). The bone density pictures are subjected to feature extraction by the DCNN, and the information is then classified into two categories using the MSVM. The National Health and Nutrition Examination Survey (NHANES) database's dataset of bone density photos was used to train and evaluate the suggested hybrid model. According to the results, the ensemble model performs better than the most advanced methods available today in terms of F1 score, sensitivity, accuracy, and specificity. According to our research, osteoporosis may be efficiently classified by the DCNN and MSVM ensemble model, which can help with the diagnosis and treatment of various bone disorders. The proposed model gives better performance in terms of accuracy of 0.8913 and specificity of 0.9123 when compared to other models. Thus, a deep-learning diagnostic network applied to lumbar spine radiographs could facilitate screening for osteoporosis.