An Extensive Survey of Deep Learning Models for Osteoporosis Detection

Abstract

Osteoporosis is a condition characterized by low bone mass and structural deterioration of bone tissue and is increasingly becoming a crucial health concern, especially among the older population. The condition also affects the skeletal system in later life. Recognition of pre-disease states during early osteoporosis management plays a critical role in preventing osteoporosis development. The present study is a comprehensive investigation concerning the establishment and assessment of osteoporosis detection deep learning models developed with data obtained from individuals over fifty years. The investigation aimed to develop and compare predictive models engineered for the purpose to classify normal and osteoporosis users. To determine the performance of the newly developed deep learning models, this study compared the profound learning methods to prediction models. The data were collected from community-dwelling participants at a medical center in Taiwan who enrolled in a health check-up program. Various images were collected to extract the features. These modalities included in the prediction model were obtained from chest CT and CT radiomics, while the extracted features from dental radiographic images were also utilized to classify osteoporosis using deep convolutional neural network models. The results of this test prove that deep learning models could significantly increase the efficiency of osteoporosis determination. Such comparative analysis validates certain deep learning concepts, such as DenseNet121 in distinguishing normal and osteoporosis user sets with a high accuracy rate. Concerning these new machine learning concepts, new insights possible persons on a greater vein at risk may be classified for preventive measures.