A deep learning-based integrated analytical system for tumor exosome on-chip isolation and automated image identification

Abstract

Exosomes are nanoscale lipid-bound vesicles secreted by various types of parent cells into the extracellular environment. They carry a wide range of bioactive molecules and serve as a crucial role in intercellular communication and tumor progression. Here, we develop an integrated microfluidic system for on-chip exosome isolation and quantum dot-based tumor marker analysis. This system integrates exosome processing and marker abundance analysis within a centimeter-scaled microfluidic chip, eliminating the need for additional off-chip treatments. We also implement YOLO v8-based image identification for sensitive and automatic detection, reducing the limit of detection (LOD) to 8.65 per microliter while minimizing manual measurement errors. Using this system, two tumor markers among four cell lines were profiled in parallel, revealing unique tumor burdens and demonstrating strong consistency with approved serological marker testing. These results highlight the potential of this technique for sensitive, precise, and automatic exosome tumor detection, paving the way for early cancer diagnosis and analysis.