Microwave bone fracture diagnosis using deep neural network

Sina Beyraghi, Fardin Ghorbani, Javad Shabanpour, Mir Emad Lajevardi, Vahid Nayyeri*, Pai Yen Chen, Omar M. Ramahi

*Corresponding author for this work

Research output: Contribution to journalArticleScientificpeer-review

9 Downloads (Pure)

Abstract

This paper studies the feasibility of a deep neural network (DNN) approach for bone fracture diagnosis based on the non-invasive propagation of radio frequency waves. In contrast to previous “semi-automated” techniques, where X-ray images were used as the network input, in this work, we use S-parameters profiles for DNN training to avoid labeling and data collection problems. Our designed network can simultaneously classify different complex fracture types (normal, transverse, oblique, and comminuted) and estimate the length of the cracks. The proposed system can be used as a portable device in ambulances, retirement houses, and low-income settings for fast preliminary diagnosis in emergency locations when expert radiologists are not available. Using accurate modeling of the human body as well as changing tissue diameters to emulate various anatomical regions, we have created our datasets. Our numerical results show that our design DNN is successfully trained without overfitting. Finally, for the validation of the numerical results, different sets of experiments have been done on the sheep femur bones covered by the liquid phantom. Experimental results demonstrate that fracture types can be correctly classified without using potentially harmful and ionizing X-rays.

Original languageEnglish
Article number16957
Number of pages11
JournalScientific Reports
Volume13
Issue number1
DOIs
Publication statusPublished - Dec 2023
MoE publication typeA1 Journal article-refereed

Fingerprint

Dive into the research topics of 'Microwave bone fracture diagnosis using deep neural network'. Together they form a unique fingerprint.

Cite this