Heavy Metal Risks in Milk
Traditional choline detection techniques such as gas chromatography-mass spectrometry (GC-MS) and high-performance liquid chromatography (HPLC) provide high sensitivity but also pose significant drawbacks, including expensive instrumentation, long assay times, and the need for skilled operators. These limitations restrict their application in resource-limited or decentralized settings where infant food quality control is vital.
Chemiluminescence (CL)-based biosensors have emerged as an attractive alternative. The core principle exploited in this study involves choline oxidase catalyzing the oxidation of choline to produce hydrogen peroxide, which reacts with luminol and cobalt ions on the sensor to emit light proportional to the choline concentration.
However, existing CL detection devices often rely on photomultiplier tubes or CCD cameras that are bulky, costly, and unsuitable for field deployment. To overcome this, integrating paper-based analytical devices (PADs) with smartphone cameras provides a low-cost, disposable platform with passive fluid handling and simple readout.
Despite progress in smartphone-based CL sensing, variations in ambient lighting and camera parameters hinder accuracy and reproducibility. Here, machine learning (ML) algorithms are introduced to address these issues by learning complex image features, normalizing environmental variations, and enabling precise quantification without conventional calibration protocols.
Synthesis and Electrode Preparation
The paper-based luminol-cobalt chemiluminescence (PLC-CL) sensor was fabricated using Whatman Grade-1 filter paper due to its porosity and capillary flow properties. Hydrophilic reaction zones and hydrophobic barriers were created using a wax-printing technique, followed by hot lamination to achieve robust fluid containment with defined sensing regions.
The biosensor was functionalized by applying an optimized concentration (3 mM) of luminol and cobalt(II) chloride, then incubated at 70°C to activate. Choline oxidase was subsequently introduced to catalyze the oxidation of choline in the sample, generating hydrogen peroxide, which triggers the luminol-cobalt CL reaction, emitting measurable light.
The assembled biosensor fit within a 3D-printed dark enclosure integrated with a smartphone module designed for CL signal acquisition via camera imaging. This platform enabled real-time detection by capturing chemiluminescence images under controlled conditions to minimize background noise. For data analysis, multiple regression models, including polynomial regression, gradient boosting, and support vector regression, were employed to predict choline concentrations from image-intensity data.
This approach eliminated the need for standard calibration curves and compensated for environmental and instrumental variability. The workflow from sample application to smartphone analysis leveraged the synergy of chemical sensing, paper microfluidics, and AI-enabled analytics.
Electrochemical Characterization and Milk Analysis
Optimization experiments refined key parameters affecting CL signal generation, including buffer pH, luminol and cobalt concentrations, reaction temperature, and sample incubation time, ensuring reproducibility and signal stability. The sensor exhibited a linear detection range for choline between 0.5 mM and 10 mM with a detection limit of approximately 257.12 μM.
This sensitivity spans physiologically relevant choline levels found in human milk and infant formulas, ranging from high micromolar to low millimolar concentrations. The biosensor displayed excellent repeatability with relative standard deviation values below 5% across multiple sensor units, confirming manufacturing consistency. Stability tests demonstrated that the device retained functionality for up to 21 days when stored appropriately, ensuring practical shelf life for field use.
Selectivity studies revealed negligible interference from common milk constituents such as glucose, lactose, calcium, urea, starch, and iodine, underscoring the sensor’s specificity to choline. The integration with a smartphone-based imaging system enabled convenient signal acquisition, while machine learning models accurately quantified choline by processing chemiluminescence images.
Download the PDF of this page here
This combined approach outperformed traditional calibration-based linear regression by adapting dynamically to variable imaging conditions, resulting in reliable, automated, and user-friendly analysis. An assembled prototype, including a syringe-based sample-loading mechanism, accommodated disposable sensing pads, highlighting the potential for real-world point-of-care deployment.
Comparative evaluation with existing choline assays emphasized that although the PLC-CL sensor has a higher detection limit compared to microscale or enzymatic sensors, its wide dynamic range and ease of use in milk samples make it well-suited for routine nutritional screening.
Validated Electrochemical Sensor for Milk Safety
The research successfully developed a low-cost, disposable paper-based chemiluminescence biosensor capable of accurately measuring choline concentrations in infant milk using a smartphone for readout. By leveraging choline oxidase, luminol-cobalt chemistry, paper microfluidics, and machine-learning-assisted data analysis, the sensor demonstrated optimal performance, with a detection limit suitable for real samples.
Ultimately, these advances aim to deliver a fully portable, smartphone-enabled analytical system with real-time, on-site food safety monitoring capabilities. This work paves the way for the democratization of diagnostics in infant nutrition, providing an accessible and scalable tool for ensuring food quality and regulatory compliance globally.
Journal reference
Zalke J.B., Kaushik V., et al. (2026). Machine learning-assisted paper-based chemiluminescence biosensor for choline quantification in infant milk: toward portable nutritional quality monitoring. Scientific Reports. DOI: 10.1038/s41598-026-50484-4, https://www.nature.com/articles/s41598-026-50484-4