Cell classification based on different off-axis holographic viewings angle using label-free imaging flow cytometry

This study presents a new method for cell classification via label-free quantitative imaging flow cytometry based on digital holography, providing insights of morphology and dynamics without the need for cell staining. By analyzing off-axis holographic images, rather than quantitative-phase-profile images, we simplify computational processes, reduce computational time and gain additional information content. In addition, we leverage multiple-viewpoint holographic projections of the cells rotated during flow to capture richer cellular information. Using convolutional neural network, we directly process each holographic projection independently through late fusion, enhancing feature characterization for each projection and achieving superior accuracy. Our model achieved a 7.69% improvement when processing ten projections compared to a single hologram and 17.95% when using multiple off-axis holographic projections compared to multiple quantitative phase profile projections. This new approach paves the way for potential real-time cell classification in imaging flow cytometry.