EventsThe 2nd International Conference on Bioengineering
Published
This submission belongs to the session S4. Biomedical Engineering and Biomaterials of the event The 2nd International Conference on Bioengineering
Published date
16 Nov, 2026
Citation
Dafner Yomira Condori, Carmen Mayorga Martinez, Machine learning-assisted multispectral platform for paper-based optical biosensors, in Proceedings of The 2nd International Conference on Bioengineering, Barcelona, 10 November–13 November 2026, MDPI: Basel, Switzerland
Share
Email
Facebook
Twitter
LinkedIn

Machine learning-assisted multispectral platform for paper-based optical biosensors

Carmen Mayorga Martinez 2
1. Biomedical Engineering Program, Faculty of Engineering, Peruvian University of Applied Sciences (UPC), Santiago de Surco, Lima, 15023
2. School of Biomedical Engineering, Peruvian University of Applied Sciences (UPC), Santiago de Surco, Lima, 15023
Abstract

Quantifying optical signals generated by paper-based enzymatic colorimetric biosensors offers a low-cost alternative for decentralized medical diagnostics. Currently, smartphone-based systems have been developed. However, the main limitations of these approaches arise from their dependence on smartphone cameras.1,2 These systems are sensitive to variations in illumination, image acquisition, and image processing. This work proposes a portable, machine-learning-assisted multispectral platform to estimate concentrations from discrete optical signals generated by colored enzymatic reactions3. For this purpose, a platform controlled by an ESP32-S3 microcontroller and comprising a six-sensor AS7262 multispectral array was developed. Each sensor detects a specific enzymatic reaction associated with a biomarker, such as glucose, urea, or creatinine. The system transmits data through a local Wi-Fi network to a web interface. These data are then exported for analysis. The resulting matrix will be used to train a supervised model based on partial least squares regression (PLS). Preliminary results for glucose detection showed that the system measures within the linear range of 2 mM to 10 mM with R=0.9719 and a sensitivity of 31.81 sensor counts/mM using the sensor's 550 nm channel. Moreover, the system shows great repeatability, with a coefficient of variation of 1.07%. The platform demonstrates the feasibility of estimating glucose concentrations.

1.    Ghateii, S.; Jahanshahi, A. Sensors Actuators B Chem. 2024, 400, 134835.

2.    Khanal, B.; Pokhrel, P.; Khanal, B.; Giri, B. ACS Omega 2021, 6, 33837–33845.

3.    Conrado, J. A. M.; de Jesus, D. M.; Machado, C. C. S.; da Mota, Y. S. N.; da Silva, S. G.; Petruci, J. F. S. Anal. Chem. 2025, 97, 20585–20593.

Keywords
Paper-based biosensor
Colorimetric sensing
Multispectral sensor
Machine learning
Biomarker quantification.
Simulation-Driven Development and Bench Testing of a Clog-Free Endoscopic Spray Device for Upper Gastrointestinal Bleeding
Development and Validation of a Low-Cost 3D-Printed Smartphone-Based Laser Fluorescence Microscope for Tuberculosis Diagnosis