Fig. 1: Overall principle and workflow of the SMHI platform.

a Schematic of the spatial hydrodynamic focusing digital holographic microscopy (DHM) imaging. In the microfluidic module, two vertical sheath flows (blue) are initially employed to focus the sample flow (gray) vertically, followed by two horizontal sheath flows (yellow) to complete the spatial focusing. By precisely controlling fluid dynamics, each cell is directly guided onto the DHM’s focal plane, finally producing in-focus single-cell holograms, which are recorded by a complementary metal-oxide-semiconductor (CMOS) camera. The main optical components of DHM include: beam splitter (BS), reflector (REFL), microscope objective (MO), tube lens (TL), and neutral density filter (NDF). b Quantitative phase microscopy (QPM) reconstruction process. It is performed on the holograms to obtain precise phase maps of single cells, including frequency filtering, numerical reconstruction, phase extraction, phase unwrapping and phase correction. c Phase feature extraction process. For each QPM phase map, the region of interest (ROI) is segmented by mask, and 81 features are extracted, including 14 bulk features, 17 first-order histogram features, and 50 high-order texture features, which are composed of 20 gray-level co-occurrence matrix (GLCM) features, 15 gray-level gradient co-occurrence matrix (GGCM) features and 15 gray-level run-length matrix (GLRLM) features. d Phase feature analysis process. It includes numerical distribution visualization, significance testing and correlation analysis to comprehensively assess the feature datasets. e Machine learning model construction process. First, the features are scored and ranked for importance using the maximun-relevance and minimun-redundancy (MRMR) algorithm. Machine learning models are then incrementally trained by adding features one by one, with classifier performance finally evaluated using confusion matrices and receiver operating characteristic (ROC) curves.