This research investigates the critical role of frequency selection in Electrical Impedance Tomography (EIT), a non-invasive imaging technique that reconstructs internal conductivity distributions through injected electrical currents. Empirical frequency selection is paramount to maximizing the fidelity and specificity of EIT images. The study explores the impact of distinct frequency ranges—low, medium, and high—on image contrast and clarity, particularly focusing on differentiating conductive materials from non-conductive materials. The findings reveal distinct empirical frequency bands for enhancing the respective contrasts: 15–38 kHz for conductive materials (copper) and 45–75 kHz for non-conductive materials (acrylic resin). These insights shed light on the frequency-dependent nature of material contrast in EIT images, guiding the selection of empirical operating ranges for various target materials. This research paves the way for improved sensitivity and broader applicability of EIT in diverse areas.

Diabetes treatment requires reliable blood glucose monitoring to allow for appropriate adjustments and interventions. The conventional finger-prick test is associated with significant discomfort and poor compliance, highlighting the critical need for non-invasive, user-friendly solutions. This study proposes and validates an open-source, multi-wavelength spectroscopy system that combines visible (660 nm) and near-infrared (940 nm) wavelengths through six distinct transmission and reflection strategies. The system was validated through a rigorous three-phase process. Phase I established a preliminary regression model against a commercial meter using 500 controlled samples (\(R^2\) up to 0.8832). Phase II expanded this validation in 96 healthy volunteers, confirming high clinical acceptability (over 95% in Zone A, Clarke Error Grid) and a mean absolute error of 8.438 mg/dL relative to the commercial device. Critically, Phase III performed a definitive clinical validation, comparing our system and the commercial meter against the venous plasma glucose (VPG) gold standard. This analysis definitively established the clinical failure of all reflection-based strategies (resulting in Zone D errors), while confirming 100% clinical safety (Zones A/B) for all transmission modes. Notably, our multi-wavelength transmission (MT) strategy achieved 100% Zone A accuracy and demonstrated superior analytical precision (tighter ISO 15197:2015 compliance) than the ISO-compliant commercial benchmark when both were compared to the VPG. This work validates a transparent, open-source platform as a clinically reliable and accurate non-invasive alternative for intermittent glucose monitoring.

In the food industry, the need to assess and predict the ripening process of fruits plays an important role in optimizing storage and transportation strategies, ensuring the product quality when reaching consumers. The study proposes a method using microwave spectrum based on vector network analyzer combined with machine learning models. This method is used to evaluate among numerous machine learning models and predict the number of days required for an unripe banana to turn to semi ripe and then ripe banana. Data is collected through scattering parameters, including reflection parameter S11 and transmission parameter S21, in the frequency range from 1 GHz to 5 GHz. The S-parameters are processed, analyzed and extracted characteristic data and fed into the machine learning models to perform the comparison and prediction process.

Electrical Impedance Tomography (EIT) is a non-invasive method for imaging conductivity distributions within a target area. The inverse problem associated with EIT is nonlinear and ill-posed, leading to low spatial resolution reconstructions. Iterative algorithms are widely employed to address complex inverse problems. However, current iterative methods have notable limitations, such as the arbitrary and subjective selection of initial parameters, lengthy computational times due to numerous iterations, and the generation of reconstructions that suffer from shape blurring and a lack of higher-order detail. To address these issues, this study investigates the impact of using ISTA and FISTA iterative algorithms on the image reconstruction process in EIT. It focuses on enhancing the convergence and accuracy of EIT image reconstruction by evaluating the effectiveness of these optimization algorithms when applied to regularized inverse problems, using standard regularization techniques. ISTA and FISTA were compared to the NOSER and Newton-Raphson methods and validated through simulation and experimental results. The results show that ISTA and FISTA achieve better visualization and faster convergence than conventional methods in terms of computational efficiency when solving regularized problems, achieving conductivity reconstructions with an accuracy of above 80%. The paper concludes that while ISTA and FISTA significantly enhance EIT image reconstruction performance, the quality of the reconstructed images heavily depends on the choice of regularization methods and parameters, which are crucial to the reconstruction process.

Non-invasive blood glucose monitoring offers a safer and more comfortable alternative to conventional invasive methods. By eliminating the need for skin puncture, it removes the associated risks of pain and potential infection. In this study, the performance of glucose prediction using the transmission method. The evaluation is conducted at three wavelengths including 660 nm, 780 nm, and 940 nm. Optical signals transmitted through the fingertip are processed to estimate blood glucose levels and validated against reference measurements. Model performance is assessed using Mean Absolute Percentage Error, the Clarke Error Grid analysis, ISO 15197:2015 and FDA standards. Among the wavelengths, the 940 nm wavelength achieves the best performance, with MAPE = 6.304% and all measurements fall within Zone A of the CEG. The 660 nm and 780 nm wavelengths have a lower accuracy, with MAPE values of 6.671% and 6.396%, respectively. Based on these results, the selection of appropriate wavelength is crucial to improving the accuracy and reliability of non-invasive glucose monitoring systems.

Physiological signal emulators support the calibration, validation, and stress testing of cardiovascular monitoring devices. However, many existing systems generate electrocardiography (ECG) or photoplethysmography (PPG) independently and offer limited control over arrhythmia detection and ECG–PPG coupling. This study presents a programmable physiological signal emulator that integrates a unified event-driven ECG–PPG model with synchronized multichannel hardware. The model represents atrial pacing, atrioventricular conduction, and ventricular activation as separate functional blocks, enabling normal sinus rhythm, first-degree atrioventricular block, second-degree atrioventricular block Mobitz I, complete atrioventricular block, atrial tachycardia, and ventricular tachycardia. A Gaussian-based ECG is generated from the atrial and ventricular event sequences, while a multi-Gaussian PPG waveform is derived from ventricular activation using a beat-class-dependent electromechanical delay. The same processing architecture supports playback of recorded 12-lead clinical ECG data through an inverse lead transformation. The hardware uses an STM32F407VET6 microcontroller and MCP4921 digital-to-analog converters (DACs) to generate 10 synchronized analog outputs, comprising 09 ECG electrodes and 01 PPG channel. Validation covered physiological timing, analog-chain performance, and end-to-end signal reproduction. PR interval errors relative to a commercial electrocardiograph were 1.23 ms for normal sinus rhythm and 1.66 ms for first-degree atrioventricular block. The measured beat-to-beat PR increment during Mobitz I conduction was $40.02\pm 0.04$ ms for a programmed value of 40 ms. At commanded amplitudes of at least 800 mV, both output channels achieved absolute amplitude errors below 0.60%, total harmonic distortion below 1%, and signal-to-noise ratios (SNRs) above 30 dB. Inter-channel R-peak skew remained below the 2 ms sampling interval, and all monitored metrics varied by less than 1.5% during 60 min of continuous operation. Reproduction of a clinical 12-lead recording yielded per-lead $R^2$ values of 0.967–0.986 and a cycle-to-cycle correlation of 0.996. The emulator also reproduced amplitude-dependent bias in automated interval measurements and interpretation labels. These results demonstrate a low-cost, open-source platform for reproducible device calibration, algorithm stress testing, medical training, and physiological signal processing research.