Cell Classification in Digital Holographic Flow Cytometry via Direct Hologram Analysis Using Deep Learning

Prostate cancer is a highly heterogeneous malignant tumor, and the classification of its cells at different stages or subtypes is of great significance. In recent years, cell classification using deep learning combined with digital holographic microscopy (DHM) has gained widespread attention. However, accurately distinguishing cells from different stages or subtypes remains a challenging task. Typically, quantitative phase images of cells are retrieved from holograms and used as inputs for classification networks. However, the phase reconstruction process is highly time-consuming and prone to errors. Here, we address the cell classification problem by directly utilizing raw digital holograms without any additional data processing. In this study, we propose a strategy for single-cell classification by integrating universal network (U-Net) with residual network-50 (ResNet-50) and residual multi-layer perceptron (ResMLP), bypassing the conventional reconstruction process to extract relevant information. This approach enables the classification of two heterogeneous prostate cancer cell lines and a type of white blood cell. The results demonstrate the effectiveness of the proposed strategy.