According to the WHO guidelines, BMD ≤ 2.5 standard deviations below the young adult mean (T-score ≤ − 2.5) indicates osteoporosis, while a T-score at any site between − 1.0 and − 2.5 indicates low bone mass or osteopenia. To date, the gold standard for osteoporosis diagnosis is the estimation of bone mineral density (BMD) in the hip and lumbar spine using dual-energy X-ray absorptiometry (DXA) 4. Hip fractures, one of the major osteoporotic fractures, are associated with limitations in ambulation, chronic pain and disability, loss of independence, and decreased quality of life, and 21%–30% of patients who have hip fracture die within 1 year 3. In the United States, the incidence of osteoporosis-related fractures is more than four times higher compared to that of stroke, heart attack, and breast cancer 1, and based on the meeting report of the World Health Organization (WHO), osteoporotic fractures account for more hospital bed-days than those diseases in several high-income countries 2. Early detection of osteoporosis is greatly important in preventing osteoporotic fractures. Osteoporosis is a common condition, especially in postmenopausal women however, it often remains undetected until after fracture occurs. The DNN model could be considered as one of the useful screening tools for easy prediction of osteoporosis in the real-world clinical setting. All Grad-CAM results from both internal and external validation sets appropriately matched the proximal femur cortex and trabecular patterns of the radiographs. The external validation set confirmed a model performance with an overall accuracy of 71.8% and an AUC value of 0.700.

The area under the ROC curve value was 0.867, indicating a reasonable performance for screening osteoporosis by simple hip radiography. The PPV was 78.5%, and the NPV was 86.1%. Our final DNN model showed an overall accuracy of 81.2%, sensitivity of 91.1%, and specificity of 68.9%. Additionally, we performed external validation using 117 datasets. A gradient-based class activation map (Grad-CAM) overlapping the original image was also used to visualize the model performance.

We drew the receiver operating characteristic (ROC) curve. We calculated the confusion matrix and evaluated the accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Based on VGG16 equipped with nonlocal neural network, we developed a deep neural network (DNN) model. The 1001 images were randomly divided into three sets: 800 images for the training, 100 images for the validation, and 101 images for the test. Of these, 504 patients had osteoporosis (T-score ≤ − 2.5), and 497 patients did not have osteoporosis. A total of 1001 datasets of proximal femur DXA with matched same-side cropped simple hip bone radiographic images of female patients aged ≥ 55 years were collected. This study aimed to predict osteoporosis via simple hip radiography using deep learning algorithm. Despite being the gold standard for diagnosis of osteoporosis, dual-energy X-ray absorptiometry (DXA) could not be widely used as a screening tool for osteoporosis.

Normal male hip xray