ECLStat: A robust machine learning based visual imaging tool for electrochemiluminescence biosensing.

Visual electrochemiluminescence (ECL) has emerged as a prominent diagnostic method for accurately quantifying various disease markers even at point of care setting with high sensitivity and accuracy. It does not employ complicated instruments such as potentiostat and expensive imaging microscopy for quantifying trace amounts of molecules. The ECL system offers significant advantages over other detection processes, such as high sensitivity, selectivity, rapid response, multiplexing, and miniaturization capabilities, making it well-suited for future commercialization. However, the current ECL system lacks standardization and accuracy in the resulting output data due to the manual measurement of ECL signal response using open-source image processing software, which often limits the efficiency of the ECL process in real-time applications. To address the shortcomings of the existing approach and advance the ECL detection process, a fully automated machine learning-assisted standalone graphical user interface (GUI) application was developed for dedicated measurement and management of ECL-emitted light signals. The working performance of the developed program is evaluated for its real-time utility by detecting hydrogen peroxide, which is an important reactive oxygen species, and glucose, which is a significant biomarker of diabetes. The obtained results show the detection limit of 0.024 mM and 0.035 mM for HO and glucose, with a quantification limit of 0.074 mM and 0.10 mM, respectively. The ultimate objective of the developed application is to improve accuracy by enabling users to apply machine learning algorithms to raw image data seamlessly without deeply comprehending the underlying computational processes and establish a standard protocol for ECL signal measurements. Moreover, the developed application can be used in other optical detection approaches such as chemiluminescence, colorimetric, and fluorescence.