Stain-free Holographic Detection of Circulating Tumor Cells Using A Deep Feature Fusion Neural Network

Circulating tumor cells (CTCs) have important reference value in cancer diagnostics. As the existing fluorescence-based CTC detection method faces the limitation of tedious staining steps and photobleaching, it's necessary to develop a stain-free identification method. This work demonstrated a stain-free holographic detection of CTCs using a deep feature fusion neural network. By utilizing the subtle difference between CTCs internal structures captured by the holographic microscope, the proposed neural network fuses low-level features extracted by the shallow layers with high-level features extracted by deep layers into fused features for CTCs identification, and reached an accuracy of 94%. Compared with the previous fluorescent stain-based method, this work provides a potential solution for high accuracy stain-free CTC detections.