Microwave Cytometry with Machine Learning for Shape-Resolved Microplastic Detection
Abstract
Microplastics are increasingly recognized as a global environmental health threat, yet their detection and characterization remain constrained by the cost, form factor, and throughput of existing analytical tools. Portable micro/nanotechnology-based sensors are emerging to address this need, but most rely on the assumption of spherical particle geometry in their operating principle, limiting their relevance for environmental analysis. Here, we overcome this limitation by advancing microwave cytometry with machine learning-enabled shape recognition. Microwave cytometry is a flow-through electronic platform that integrates microwave resonator responses with low-frequency impedance signals to capture the dielectric signatures of individual particles. Using microscopy-derived shape measurements as ground truth, we trained a random forest model to decode these information-rich waveforms. Once trained, the system operates without optical input, enabling electronic-only determination of particle geometry. We demonstrate extraction of the major and minor axes of ellipsoidal microparticles with <8% relative error on average and use these predictions to derive the dielectric permittivity of ellipsoid particles. This approach removes long-standing shape assumptions in microplastic sensing and establishes a pathway toward portable, high-throughput, morphology-aware detection technologies.