Deep learning–enabled investigation on non-invasive glucose detection using miniaturized pentagonal slotted microstrip patch antenna (MPSMA)

Abstract A miniaturized pentagonal slotted microstrip patch antenna (MPSMA) is designed using FR4 substrate at 4.2 GHz for non-invasive glucose sensing applications. The MPSMA design was optimized for maximal impedance matching, S 11 , Voltage Standing Wave Ratio (VSWR), efficiency, gain, and directivity. Simulated results were experimentally validated using a vector network analyzer inside an anechoic chamber. The proposed MPSMA achieved a notable return loss of −36.36 dB, a VSWR of 1.0308, efficiency of 75.4 %, gain of 1.5 dBi, and directivity of 2.879 dBi, confirming its performance. A detailed analysis was carried out by placing the antenna near various concentrations of glucose solution (10%–50%), which indicates a significant shift in the resonant frequency and reflection characteristics of the antenna. Principal component analysis was applied for identifying distinct characteristics between various glucose concentrations. The glucose sensing capability of the proposed MPSMA was validated by using a multilayer finger model designed in CST Studio software that mimics real tissue characteristics (Cole–Cole model). The detailed analysis on the multilayer finger model as well as the beaker solution setup highlights the potential of MPSMA in non-invasive glucose monitoring. Further deep learning is implemented using the experimental data for the prediction of glucose from the S 11 data. Convolutional neural network classifier results are balanced across most classes. Around most glucose concentrations, true and predicted values are close, with minor overestimation around 4000 mg/dL. The proposed solution is inexpensive and has the potential for diabetes treatment, which ensures the feasibility integration of microstrip antenna technology into next-generation biomedical sensing platforms.