Three-dimensional forward-scattering imaging flow cytometry system for single-cell analysis Authors: Minhong Zhou, Jingjing Zhao, Xinyu Chen, Zunming Zhang, Zhaoyu Lai, Ziqi Zhou, Adam de la Zerda, Yu-Hwa Lo Published: 2025 ABSTRACT Label-free single-cell 3D imaging is essential for accurate phenotyping by minimizing perturbations to native cellular states and facilitating downstream molecular analyses. Among the available modalities, 3D forward-scattering (dark-field) imaging offers the richest subcellular details but faces challenges from strong transmitted-beam interference. We present a high-throughput 3D forward-scattering imaging flow cytometry system employing optical needle-beam illumination, linear micro-mirror arrays for axial scatter detection, and spatiotemporal deconvolution algorithms. Experimental validations with microstructures, hydrogel beads, and HEK-293 cells confirm the method’s capability for robust subcellular resolution at ∼400 cells/s, enabling advanced label-free cellular diagnostics and analysis. This platform enables label-free, high-content cellular analysis and is readily adaptable to integrated cell sorting and AI-driven classification, offering broad potential for biomedical research and diagnostics. INTRODUCTION Label-free single-cell imaging has become pivotal for precise cell phenotyping, enabling the correlation of morphological features with underlying functional and molecular characteristics without perturbing native cellular states.1,2 Unlike labeled imaging techniques, label-free modalities preserve intrinsic cellular properties, facilitating subsequent multi-omics analyses and clinical diagnostics.3–5 Imaging flow cytometry (IFC) uniquely provides high-resolution images of single cells at subcellular detail in a high-throughput manner, thereby bridging traditional cytometric assays and microscopy.6–9 However, current IFC systems predominantly deliver two-dimensional (2D) information,10,11 limiting their ability to accurately capture spatial complexity. Critical intracellular structures and interactions may be obscured or inaccurately interpreted when projected onto a 2D plane. For instance, distinct scattering centers in three-dimensional (3D) space can appear overlapped in a single-plane projection, leading to erroneous biological conclusions. Conversely, 3D imaging techniques significantly enhance spatial resolution and reveal dynamic intracellular processes critical for cell biology and biomedical research.12,13 Recent progress in 3D IFC has centered on fluorescence tomography,14–17 powering AI-driven classification, subtyping, and fate prediction. For example, Hua et al. achieved multi-color 3D fluorescence imaging at up to 5750 cells/s.18 However, these fluorescence images require labels and pre-treatment of samples, and these processes may disturb cells and create bias for phenotypes. Pirone et al. presented a stain-free strategy that combines tomographic phase microscopy with computational statistical-inference segmentation (CSSI) to identify nuclei,19 but the approach depends on models trained on fluorescence ground truth and operates at low throughput (∼1 to 2 cells/s). Building on side-scattering tomography and spatiotemporal mapping, our group previously reported a 3D side-scattering imaging flow-cytometry platform20,21 and showed that 3D scattering encoded rich information for cell classification, type discovery, and fate prediction.22 In this paper, we incorporate the design of micro-mirror array z-microscopy23 into a microfluidic flow cytometer platform to realize high-throughput, label-free 3D forward-scattering (FSC) single-cell imaging. Our solution addresses previous limitations by employing three core innovations: (a) needle-shaped laser illumination to reduce background noise; (b) linear micro-mirror arrays coupled with a photomultiplier tube (PMT) array to achieve axial resolution; and (c) a sophisticated spatiotemporal deconvolution algorithm to reconstruct high-contrast volumetric images from time-resolved signals. The system is broadly compatible with widely used microfluidic chips and cartridges. Validated with microfabricated structures, hydrogel cell simulants, and biological cells (HEK-293), our platform achieves robust subcellular resolution at ∼400 cells per second, establishing a foundation for enhanced diagnostic capabilities, integrated image-activated sorting, and AI-driven cellular analysis. PRINCIPLE AND METHODS Figure 1(a) presents a simplified schematic of the 3D-FSC-IFC system. The 3D-FSC-IFC system employs a 50 mW, 488 nm laser (SF NX 488-50 CDRH, Sapphire). The laser polarization direction is changed from horizontal to vertical by a half-wave plate (WPH10M-488, Thorlabs) before being deflected by an acousto-optic deflector (AOD) (OAD948, ISOMET) driven by a sawtooth voltage signal from a waveform generator (WaveStation 2022, Teledyne LeCroy). This setup creates a scanning laser beam at 200 kHz. A 4f system with two lenses (L1: f = 40 mm, LBF254-040-A, Thorlabs; L2: f = 150 mm, LBF254-150-A, Thorlabs) is used to expand the beam. A 50:50 beam splitter (BS013, Thorlabs) is used to introduce the LED (M455L3, Thorlabs) into the setup to aid visualization and optical alignment. A customized diffractive optical element (DOE) is placed before the input objective (IO) to generate a needle beam. A 20×/0.42 Plan Apo Infinity-Corrected Long WD objective lens (Mitutoyo) was used for illumination. A 50×/0.55 Plan Apo Infinity-Corrected Long WD objective lens (Mitutoyo) is used for detection. The detection objective is mounted on a translation stage to optimize the working distance for different operating conditions. Emergent light is split by a 10:90 beam splitter (BS025, Thorlabs), 10% through lens L3 (LBF254-200-A, Thorlabs) to the camera (CS165CU, Thorlabs) to help focusing and 90% through lens L4 (LBF254-200-A, Thorlabs) to the micro-mirror array. Each micro-mirror in the mirror array is 70 μm high and 5 mm long, fabricated by 3D printing (Nanoscribe Photonics GT) and coated with a 200 nm aluminum layer by sputtering to enhance reflectivity. A 3D printed spatial filter is placed before L4 to block transmission light. Micro-mirror array geometry and objective selection are determined by resolution and fabrication requirements. Full specifications are given in the supplementary material. Implementation of the 3D-FSC-IFC system. (a) Schematic diagram of the 3D-FSC-IFC system. AOD: acousto-optic deflector; DOE: diffractive optical element; IO: 20X/0.42 illumination objective lens; DO: 50X/0.55 detection objective lens; SF: spatial filter; BSs: beam splitters; and PMT array: eight-channel photomultiplier tube array. The AOD produces a scanning laser beam, and the DOE generates a needle-shaped beam. The sample is 2D hydrodynamically focused by the sheath flow in a microfluidic cartridge. (b) Illustration of the z-resolution principle: by blocking transmitted light, the spatial filter causes scattered light from different z-positions to focus onto distinct micro-mirror elements and separate PMT channels. (c) 3D reconstructed space. The resolution along the x-axis is determined by the laser scanning range; the resolution along the y-axis is determined by the distance the cell travels between two scans; and the resolution along the z-axis is determined by the number of PMTs and their focal plane spacing. Each time the laser illuminates an object, a 1D intensity profile along the z-axis is recorded. (d) A single laser scanning period produces a 2D xz-plane intensity profile. (e) Once an object fully passes through the scanning area, the time-domain signal contains the complete 3D profile. The design of the mirror array is described in the supplementary material. A 4f system is used to focus the forward scatter signal from the mirror array into the photomultiplier tube (PMT) array (H9520-20, Hamamatsu). The data acquisition system and the diffractive optical element are also described in the supplementary material. Cells are delivered via a disposable microfluidic cartridge (NanoCellect) mounted on a translation stage for precise positioning, with flow driven by a fluidic pump system (NanoCellect). Suspended cells are hydrodynamically focused into a single file within the channel. We operate at a mean flow speed of 0.16 m s−1, which yields robust one-dimensional hydrodynamic focusing. Under these conditions, the sample volumetric flow is 12 μl min−1, and the input cell concentration is limited to ≤2000 cells/μl to prevent aggregation and clogging. The scanning laser beam at a scanning rate of 200 KHz is a needle-shaped beam formed by a diffractive optical element (DOE), with a diffractive-limited waist of ∼3 μm and a depth of focus of ∼40 μm.24 While the laser is rated at 50 mW, the output power for typical operation is ∼3 mW. With a 50:50 beam splitter placed before the sample path, the actual laser power incident on cells is <1.5 mW. At a flow speed of 0.16 m/s, each cell resides in the scanning region for <0.2 ms, corresponding to an exposure of <0.3 μJ per cell. This low dose is well below reported photothermal and photochemical thresholds and, therefore, does not perturb cell viability or physiology.25 To deconvolute the image signal, we consider that the needle-shaped laser beam maintains a Gaussian profile with a beam width w; thus, the time-domain signal recorded by the PMT array can be expressed asPi⁢t=∑j∬x,ydxdyMij⁢y⁢exp⁡−2y2w2×fx,vy⁢t,zjexp⁡−x−xnt22w2,(1)where Pit denotes the signal recorded by the ith PMT in the array and fx,y,zj with yt=vyt represents the scattered light intensity from a traveling object at position x,y,zj at time t, in which zj is the focal position in the object space corresponding to the jth PMT. The flow speed of the object is vy, while xnt is the laser’s x position at time t during scanning, described asxnt=vxt−T2−nT,nT