Related Experiment Video
Updated: Jun 27, 2025
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Predicting osteoporosis from kidney-ureter-bladder radiographs utilizing deep convolutional neural networks
Tzu-Yun Yen1, Chan-Shien Ho1, Yu-Cheng Pei2
1Department of Physical Medicine and Rehabilitation, Chang Gung Memorial Hospital at Linkou No. 5, Fuxing Street, Guishan District, Taoyuan City 333, Taiwan; School of Medicine, Chang Gung University, No. 259, Wenhua 1st Road, Guishan District, Taoyuan City 33302, Taiwan.
This study introduces DeepDXA-KUB, a deep learning model using kidney-ureter-bladder (KUB) radiographs to predict bone mineral density (BMD) for osteoporosis screening. The model shows promising accuracy, offering a potential low-cost alternative to dual-energy X-ray absorptiometry (DXA).
Area of Science:
- Radiology
- Artificial Intelligence
- Osteoporosis Research
Background:
- Osteoporosis poses significant health risks, including fractures and mortality.
- Dual-energy X-ray absorptiometry (DXA) is the standard for diagnosis but is costly and inaccessible.
- Kidney-ureter-bladder (KUB) radiographs are common, inexpensive, and widely available, presenting an opportunity for screening.
Purpose of the Study:
- To explore the prediction of bone mineral density (BMD) using KUB radiographs.
- To develop and evaluate a deep learning model (DeepDXA-KUB) for opportunistic osteoporosis screening.
- To assess the model's ability to classify high-risk patient groups based on KUB images.
Main Methods:
- A deep learning model, DeepDXA-KUB, was developed to predict BMD from KUB radiographs.
- Utilized a dataset of 8913 KUB/DXA pairs from Taiwanese medical centers (2006-2019).
- Evaluated model performance using standard metrics including accuracy, sensitivity, specificity, and AUROC.
Main Results:
- DeepDXA-KUB demonstrated moderate correlations between predicted and DXA-measured BMD (r=0.858 lumbar, r=0.87 hip).
- Achieved high accuracy (84.7% lumbar, 84.2% hip), sensitivity (81.6% lumbar, 91.2% hip), and specificity (86.6% lumbar, 81% hip).
- Area under the receiver operating curve (AUROC) values were high (0.939 lumbar, 0.947 hip), indicating strong performance.
Conclusions:
- This is the first study to use deep learning on KUB radiographs for predicting lumbar spine and femoral BMD.
- The DeepDXA-KUB model shows significant potential for opportunistic osteoporosis screening.
- KUB radiographs can serve as a cost-effective tool for identifying individuals at risk for osteoporosis.
