Smart microfluidic device for automated exosome quantification and classification using deep learning

Microfluidic resistive pulse sensing offers a streamlined alternative for detecting and analyzing microparticles across various fields, including environmental science, chemistry, biomedicine, and disease diagnostics. This study introduces a novel microfluidic platform designed to analyze and characterize the physical properties of cancer-derived exosomes through automated, efficient analysis. The platform employs capillary- and vacuum-chamber-assisted passive-driven fluid injection, enabling fully automated operation. To ensure precise measurements, the platform incorporates hydrodynamic particle focusing without the need for a sheath and utilizes a reference gate to minimize noise during resistive pulse detection. The proposed microfluidic chip accurately measures the size, concentration, and zeta potential of exosomes derived from MCF-7, MDA-MB-231, and MCF-10A cell lines with high sensitivity. Furthermore, the microfluidic platform is fabricated through a cost-effective soft lithography method. Moreover, a deep learning-based model is applied to classify and distinguish exosomes based on the collected data, achieving an impressive accuracy of 96.6%. The microfluidic platform combines high sensitivity with advanced classification capabilities. Moreover, the proposed microfluidic devices’ compact and minimal operation advances Lab on a Chip for practical applications. • We presented smart microfluidic device for in situ analysis of breast cancer derived exosomes by resistive pulse sensing. • The microfluidic device in this work does not require external pump for fluid injection by application of passive-driven microfluidics. • The microfluidic platform analyzed physical properties (size, concentration, zeta potential) of three types of exosomes derived from two different breast cancer cell line (MCF-7, MDA-MB-231) and normal breast cell line (MCF-10A) with high sensitivity. • Based on the measured signal, the classification of three different exosomes were successfully performed by application of deep learning model with accuracy over 96%. • The microfluidic platform in this work shows great potential for miniaturized, high-throughput screening and can be useful platform for analyzing nanoparticles in various industries.