Diagnosing osteoporosis using a deep-radiomics-based approach

Researchers have developed deep-radiomics-based models capable of predicting osteoporosis with high accuracy and without need for DXA imaging. 

Their work was published recently in Radiology: Artificial Intelligence. In the paper, experts detail how they extracted radiomics features from nearly 5,000 hip radiographs, which were used to train multiple models to detect minute signs of osteoporosis.  

Although bone mineral density scans are the reference standard for diagnosing osteoporosis, they continue to be underutilized and may not be available at all in lesser developed countries, authors of the study pointed out. 

“Its use for screening and posttreatment follow-up in regions with poor economies is limited due to the low availability of scanners and relatively high cost. Even in developed countries, many patients are left at risk for fracture without undergoing osteoporosis screening due to lack of understanding among health care providers,” corresponding author Hee Dong Chae, from the Department of Radiology at Seoul National University Hospital, and co-authors explained. 

For these reasons, the utility of deep learning radiomics-based approaches to assist in diagnosis using routine radiographs has been explored. Though promising, studies pertaining to this approach have been limited in their feature extractions and comparisons.  

For this study, researchers used a multitude of feature extractions to train the deep-radiomics-based models—10 deep features, 16 texture features and three clinical features. In total, seven models were developed, each of which used different combinations of features to diagnose osteoporosis. Area under the receiver operating characteristic curve (AUC) was used grade diagnostic performance and 6 radiologist completed observer agreement tests.