Machine learning-based dual-modality rapid in-flow microplastic sizing using combined dielectrophoretic and optical features

Accurate sizing of individual microparticles within a heterogeneous sample is challenging. We present a dual-modality microfluidic platform that integrates lens-free forward scattering and in-flow dielectrophoresis (DEP) for high-throughput, rapid sizing of microparticles. Particles are injected and flow in parallel through a wide microfluidic channel. Amplitude-based features from optical diffraction patterns are recorded with a CMOS sensor array. These features are mapped to particle diameter using a machine-learning regression model trained exclusively by Mie-scattering simulated data. Mie simulations enable the generation of supervised training labels without the need for size calibration experiments. Simultaneously, the same particles are characterized by their DEP response in a non-uniform electric-field actuation region within the channel. The DEP response is quantified by measuring their velocity before and after the actuation region. A fluid dynamics simulation is used to map the DEP response to size. Experiments on monodisperse polystyrene spheres yield optical diameter distributions matching manufacturer specifications. In a poly-disperse sample of two bead sizes, optical analysis shows a bimodal size distribution, and velocity analysis shows two size-dependent clusters. Using either modality alone, of events cannot be classified, whereas combining optical and DEP modalities, the number of unclassified events is reduced to . The results demonstrate that optical diffraction and DEP analysis provide complementary, size-sensitive information for enhanced particle sizing while in flow. Our device provides a path for improved microplastics and biological single-cell analysis.