Machine learning-driven optical microfiltration device for improved nanoplastic sampling and detection in water systems

Abstract

The rising presence of nanoplastics in water poses toxicity risks and long-term ecological and health impacts. Detecting nanoplastics remains challenging due to their small size, complex chemistry, and environmental interference. Traditional filtration combined with Raman spectroscopy is time-consuming, labor-intensive, and often lacks accuracy and sensitivity. This study presents an agarose-based microfiltration device integrated with machine learning–assisted Raman analysis for nanoplastic capture and identification. The 1 % agarose microfluidic channel features circular micropost arrays enabling dual filtration: nanoplastics diffuse into the porous matrix, while larger particles (>1000 nm) are blocked by the microposts. Unlike conventional systems, this design achieves both physical separation and preconcentration, enhancing nanoplastic detectability. Upon dehydration, the agarose forms a transparent film, significantly improving Raman compatibility by minimizing background interference. This transformation enables direct Raman analysis of retained nanoparticles with enhanced signal clarity and sensitivity. Using 100-nm polystyrene nanoparticles (PSNPs) as a model, we evaluated device performance in distilled water and seawater across concentrations (6.25–50 µg/mL) and flow rates (2.5–100 µL/min). Maximum capture efficiencies of 80 % (seawater) and 66 % (distilled water) were achieved at 2.5 µL/min. A convolutional neural network (CNN) further enhanced spectral analysis, reducing mapping time by 50 % and enabling PSNP detection in seawater at 6.25 µg/mL. This agarose-based system offers a scalable, cost-effective platform for nanoplastic sampling, demonstrating the potential of combining microfluidics with machine learning–assisted Raman spectroscopy to address critical environmental and public health challenges.