Diseño y construcción de un dispositivo glucómetro IoT no invasivo y análisis comparativo con un glucómetro comercial

Abstract

CONDITION FOR PUBLICATION OF PROJECT. This project proposes the development of a non-invasive glucose measurement device, based on photoplethysmography technology and artificial intelligence algorithms, enable blood glucose monitoring without puncture. The hypothesis states that the analysis of optical signals, combined with the use of artificial neural networks, can estimate glucose levels within an acceptable error margin. The justification lies in the need for less painful and more accessible monitoring methods for both healthy individuals and people with diabetes. During the development phase, an ESP32 microcontroller was used along with infrared optical sensors, an OLED display, and signal conditioning circuits. Techniques for detecting peaks (maximum voltages) and valleys (minimum voltages) in the photoplethysmographic signal were implemented to calculate absorbance and peak-to-peak voltage, which served as inputs for a neural network trained using Google Colab. Validation standards such as ISO 15197:2015 were applied to compare the device’s performance with a commercial Accu-Chek Instant glucometer. The results showed that the device was able to estimate glucose levels with consistent accuracy within the margins allowed by applicable regulations. Linear regression yielded an average error of 5.481%, calculated from three measurements with individual errors of 5.481%, 5.396%, and 5.190%. Artificial neural networks recorded an average error of 5.926%, with individual errors of 5.740%, 6.267%, and 5.773%. These error margins are considered acceptable compared to measurements taken with a commercial glucometer. It is concluded that the developed device represents a viable basis for the development of noninvasive glucometers, combining affordable hardware with intelligent processing, and opens the door for future improvements in accuracy and portability. Keywords: Noninvasive glucometer, photoplethysmography, artificial neural networks, ESP32, glucose estimation.