Closed-Bipolar Mini Electrochemiluminescence Sensor to Detect Various Biomarkers: A Machine Learning Approach

Real-world usage of electrochemiluminescence (ECL) sensors are constrained by challenges like nonlinearity, sensor-to-sensor output variations, and multidimensionality. Machine learning (ML) can help resolve these challenges effectively. This study used closed ECL systems with luminol/H2O2-based electrochemistry to accurately measure the concentration of biomarkers such as cholesterol, choline, lactate, and glucose. A smartphone-based ECL detection for cholesterol, choline, lactate, and glucose was carried out by achieving a linear range from 0.5 to 10 mM, 0.01 to 1 mM, 0.1 to 5 mM, and 0.5 to 10 mM with limit of detection (LoD) values of 0.49, 0.01, 0.09, and 0.3 mM, respectively. Moreover, to prove the practical functionality of the ECL device, an anti-interference capability, stability, and reproducibility analysis was done. In addition, the smartphone assisted with ML approach was introduced to fasten ECL imaging. Various regression ML models (ordinary least-square regression, Huber regression, random sample consensus (RANSAC) regression, and Theil-Sen regression) were used to predict biomarker concentration and to improve accuracy. Finally, real blood serum analysis was carried out and achieved encouraging results. Based on the quantitative analytical performance with the inclusion of ML, the ECL device has the potential to be used for real-world applications.