Decoding Liquid Biopsy with AI: Freeze–Thaw-Induced Fingerprints in Small Extracellular Vesicles

Abstract

Liquid biopsy enables noninvasive cancer diagnosis via the detection of circulating tumor cells and small extracellular vesicles (sEVs), yet accurate tumor subtype discrimination remains limited by low biomarker abundance. Here, we propose a low-cost, automated cancer classification platform based on freeze–thaw-induced floating patterns of gold nanoparticles (FTFPA), integrating smartphone-based image capture and AI-driven analysis. The system classifies nine cell types and their sEVs with F1 scores of 0.891 and 0.898 (n = 864) and achieves 0.814 (n = 576) on clinical samples including healthy controls, breast nodules, and breast cancer subtypes. Capable of processing 96 samples in 1.5 min at 1% of conventional microscopy cost, the method exploits AuNP aggregation driven by freeze–induced concentration and weak interactions. This portable and rapid approach enables robust sEV classification and tumor subtype diagnosis, providing a practical solution for point-of-care cancer diagnostics.