Dual-Antiphase Patch Antennas for Microwave Imaging and Osteoporosis Screening Results Based on Neural Networks: Theoretical and Experimental Results
In this thesis, we introduce, construct, and test novel miniaturized antennas for microwave imaging. We also present the corresponding osteoporosis screening and detection results which are based on deep learning.
Microwave Imaging is an emerging medical imaging technology with potential benefits in physical size, complexity, and cost when compared to traditional solutions. There is potential to further optimize the cost of a diagnostic solution by reducing image resolution and relying on signal processing techniques to make up the difference. The required image resolution depends on the application in question. Additionally, for generic imaging, there is potential to increase image resolution using smaller antennas and higher operating frequencies that can be realized using more efficient on-body antennas.
A point-to-point transmission setup has been used to measure subjects to determine if they are osteoporotic or healthy. This setup is safe, easy to use, and compact when compared to the standard, x-ray-based imaging modality for osteoporosis. Two dichotomous diagnostic tests were performed using the subset of the study participants who could be conclusively classified as osteoporotic, osteopenic, or healthy. The first test investigated an integral-based classifier that achieved a Youden’s J index of 81.5%. The second test investigated the use of a perceptron neural network classifier that produced a Youden’s J of approximately 83%. The neural network achieved 94% specificity, making it more suitable for pre-screening potentially osteoporotic patients compared to the less specific integral classifier.
The dual antiphase patch antenna used for osteoporosis detection is inherently more efficient at radiating into the body than contemporary on-body dipole or single-patch antennas. A miniature, 2.4 GHz, version of the dual antiphase patch antenna has been developed using computer simulation, fabricated, and tested for viability in a theoretical high-resolution brain-imaging setup. The balun and matching circuitry have been condensed into the antenna’s PCB. The effect of surface waves was also factored into the design consideration, while maximizing the detected signal’s SNR.
This thesis contains two main contributions. First, a novel neural network classifier (topology, training, verification) for microwave imaging of healthy vs. osteoporotic bone based on wrist testing results. Second, the design, construction and testing of a novel miniaturized antenna for microwave imaging: a 2.4 GHz dual antiphase antenna.
Dr. Sergey Makaroff
ECE Department, WPI
Dr. Gregory M. Noetscher
ECE Department, WPI
Dr. Ara Nazarian
Musculoskeletal Translational Innovation Initiative, Carl J. Shapiro Department of Orthopedic Surgery, Beth Israel Deaconess Medical Center and Harvard Medical School
Dr. James E. Brown
Micro Systems Engineering, Inc.
Dr. Kyoko Fujimoto