AI-Driven Label-Free Exfoliated Tumor Cells Identification Using Hyperspectral Single-Cell Optofluidics

Bladder cancer poses a significant global health challenge due to its high recurrence rate and the limitations of current diagnostic methods. While cystoscopy remains the clinical gold standard, it is invasive and costly, whereas non-invasive alternatives such as urine cytology suffer from low sensitivity, especially in early-stage tumors. Therefore, this paper presents a label-free platform integrating hyperspectral imaging (HSI) with digital microfluidics to overcome the reliance on biomarkers. A deep learning model was implemented to achieve accurate sorting of exfoliated tumor cells in urine(UTCs) from endothelial progenitor cells (EPCs). Biophysical validation confirmed distinct nanoscale structural heterogeneity in UTCs, characterized by local refractive index variations and spatial disorder, demonstrating the platform's unique capability to leverage intrinsic biophysical properties for cancer cell identification without the need for labeling.