Machine Learning-Driven Optimization of Viscoelastic Microfluidic Particle Separation

Viscoelastic microfluidics (VEM) offers superior biological particle isolation but is hindered by complex dynamics necessitating laborious empirical optimization. To address this challenge, we propose a machine learning (ML)-driven strategy for the rapid identification of optimal operating conditions. By leveraging an experimentally calibrated theoretical model, we generated large-scale particle motion data sets to train a Random Forest algorithm, establishing a robust bidirectional mapping between sorting-relevant dynamic parameters and external input variables. Furthermore, Pareto optimization was integrated to elucidate intrinsic trade-offs among competing objectives, enabling the direct determination of optimal parameters tailored to specific separation requirements. This framework successfully transitions VEM-based isolation from an experience-dependent to a data-driven paradigm. By streamlining the optimization process, our strategy significantly enhances the scalability and utility of VEM for critical clinical applications.