Deep learning-enhanced study of single extracellular vesicle trapping dynamics with double-nanohole optical tweezers

Extracellular vesicles (EVs) are lipid-based structures crucial in cellular communication and disease pathology, making them valuable for non-invasive cancer diagnostics. Traditional detection methods face challenges like background fluorescence and nonspecific labeling, limiting their efficacy. This study presents a label-free framework utilizing double nanohole (DNH) optical tweezers for high-precision single EV analysis, enabling the classification of EVs from non-malignant, non-invasive, and invasive cancerous cell lines. Leveraging TrapNet, an interpretable deep learning model, we achieve nearly 99% classification accuracy from scattering signals of individually trapped EVs. This approach enhances early cancer detection sensitivity and specificity, offering a pathway for real-time, minimally invasive diagnostics and drug delivery applications.