Osteoporosis Diagnosis Based on Ultrasound Radio Frequency Signal via Multi-channel Convolutional Neural Network

• Published in: 2021 43rd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) Vol. 2021; pp. 832 - 835
• Main Authors: Chen, Zhiwei, Luo, Wenqiang, Zhang, Qi, Lei, Baiying, Wang, Tianfu, Chen, Zhong, Fu, Yuan, Guo, Peidong, Li, Changchuan, Ma, Teng, Ding, Yue, Liu, Jiang
• Format: Conference Proceeding Journal Article
• Language: English
• Published: United States IEEE 01.11.2021
• Subjects: Absorptiometry, Photon; Bones; Deep learning; Feature extraction; Humans; Neural Networks, Computer; Osteoporosis; Osteoporosis - diagnostic imaging; Radio frequency; Radio Waves; RF signals; Ultrasonic imaging; Ultrasonography
• ISSN: 2694-0604
• DOI: 10.1109/EMBC46164.2021.9629546

Osteoporosis is a metabolic osteopathy syndrome, and the incidence of osteoporosis increases significantly with age. Currently, bone quantitative ultrasound (QUS) has been considered as a potential method for screening and diagnosing osteoporosis. However, its diagnostic accuracy is quite low. By contrast, deep learning based methods have shown the great power for extracting the most discriminative features from complex data. To improve the osteoporosis diagnostic accuracy and take advantages of QUS, we devise a deep learning method based on ultrasound radio frequency (RF) signal. Specifically, we construct a multi-channel convolutional neural network (MCNN) combined with a sliding window scheme, which can enhance the number of data as well. By using speed of sound (SOS), the quantitative experimental results of our preliminary study indicate that our proposed osteoporosis diagnosis method outperforms the conventional ultrasound methods, which may assist the clinician for osteoporosis screening.