Frequency-Domain Photobleaching Denoising for Sensitive Extracellular Vesicle Detection Using Fluorescence Microscopy

Extracellular vesicles (EVs) are promising biomarkers for non-invasive cancer diagnostics, but their nanoscale size and correspondingly weak fluorescence signals make single-particle detection challenging without specialized, high-cost instrumentation. Here, we present a fast Fourier transform (FFT)-based computational imaging approach for EV analysis, termed EV-FFT, which repurposes fluorophore photobleaching into a frequency-domain signature for robust single-nanoparticle detection. EV-FFT transforms time-resolved fluorescence trajectories into Fourier space to suppress high-frequency stochastic noise while retaining the slow photobleaching component characteristic of nanoscale emitters. We validated EV-FFT using EVs from multiple cancer cell lines across eight protein markers, with signals showing strong concordance with ELISA measurements. In a proof-of-concept breast-cancer cohort, an eight-marker EV-FFT panel distinguished patients (stage II/IV) from healthy controls, achieving 97.2% sensitivity and 99.3% specificity at the Youden-index-optimized threshold (bootstrap 95% confidence intervals: 84.1-100% and 87.5-100% for sensitivity and specificity, respectively), outperforming conventional single-frame imaging (78.9%/91.1%). By improving sensitivity through computation rather than hardware complexity, EV-FFT enables accessible, high-precision EV immunophenotyping for cancer classification.