DNG metamaterial-inspired slotted Vivaldi antenna development integrated with supervised machine learning for ex-vivo bone fracture diagnosis
1
Issued Date
2025-08-01
Resource Type
ISSN
20904479
Scopus ID
2-s2.0-105005252404
Journal Title
Ain Shams Engineering Journal
Volume
16
Issue
8
Rights Holder(s)
SCOPUS
Bibliographic Citation
Ain Shams Engineering Journal Vol.16 No.8 (2025)
Suggested Citation
Hossen M.S., Hoque A., Islam M.T., Kirawanich P., Baharuddin M.H., Soliman M.S. DNG metamaterial-inspired slotted Vivaldi antenna development integrated with supervised machine learning for ex-vivo bone fracture diagnosis. Ain Shams Engineering Journal Vol.16 No.8 (2025). doi:10.1016/j.asej.2025.103464 Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/110325
Title
DNG metamaterial-inspired slotted Vivaldi antenna development integrated with supervised machine learning for ex-vivo bone fracture diagnosis
Author's Affiliation
Corresponding Author(s)
Other Contributor(s)
Abstract
The following research paper describes an Ex vivo bone fracture diagnosis method based on supervised machine learning (ML) using metamaterial-loaded slotted vivaldi antenna numerical parameters (S11). The proposed 26 mm x 14.05 mm compact antenna (operating at 6.89 GHz) is designed and prototyped using commercially available Rogers RT5880 substrate material with a standard substrate height of 0.508 mm (tanδ = 0.009, ϵr = 2.2). The double negative (DNG) metamaterial structure dramatically enhances antenna realized gain from 2.4 dB to 4.1 dB, and notable changes have been noticed in the S11, surface magnitude current distribution, specific absorption rate (SAR) distribution (both 1 g and 10 g), as well as in antenna radiation pattern properties (E field, H field) both in simulation and measurement. Efficiency curve was recorded over 85% in the operating frequency band. A cylindrical bone phantom model, exported from CST MWS, is used to collect the bone penetration S11 data for the ML analysis. 1440 measured data points have been accumulated from different fracture types to run the well-known SVM, adaptive boosting, random forest, XGBoost, decision tree, and logistic regression classifier for supervised learning. Training is carried out to ensure robust model performance with 75% to 85% of data from the dataset. The measured performance metrics and result comparison findings show that predictive models achieve good accuracy, which verifies the frequency-dependent pattern of the dataset and successfully predicts different classes of bone fracture (i.e., transverse, oblique, and green stick), which achieves the primary goal of this research.