Introduction
In clinical diagnostics and drug discovery, high-throughput analysis of cellular populations is critical. While flow cytometry is the standard for 2D analysis, capturing the internal 3D structure of cells—such as nuclear morphology or organelle distribution—without toxic labels remains a challenge. In-flow Holographic Tomography (FHT) addresses this by utilizing the natural rotation of cells in microfluidic channels to capture multiple viewing angles. However, traditional FHT algorithms are limited by the assumption of ideal, single-axis rotation. Irregularly shaped cells or complex tumbling dynamics often lead to "pose errors," forcing researchers to discard significant portions of data and introducing sampling bias.
Methodology: The OmniFHT Framework
OmniFHT is a "pose-free" 3D imaging framework designed to reconstruct high-resolution volumetric structures from cells undergoing unpredictable motion. Unlike conventional methods, OmniFHT treats reconstruction as a joint optimization problem, simultaneously estimating the cell's 3D refractive index (RI) distribution and its unknown rotational trajectory (pose).
The framework consists of several key innovations:
- Implicit Neural Representation (INR): Instead of a voxel-based grid, OmniFHT employs a Multi-Layer Perceptron (MLP) to represent the cell's scattering potential in the Fourier domain. This continuous representation acts as a regularizer, effectively mitigating the "missing wedge" problem common in tomography.
- Coarse-to-Fine Pose Searching: The system performs a hierarchical search across the $SO(3)$ rotation and $\mathbb{R}^2$ translation space for each captured frame, iteratively refining the pose until the predicted projection matches the experimental data.
- Physics-Informed Learning: The model incorporates the Fourier Diffraction Theorem and the Rytov weak-scattering approximation to maintain physical consistency between the 3D RI distribution and the measured 2D phase shifts.
Experimental Validation and Results
The researchers validated OmniFHT using synthetic datasets, standard cell lines (SW780), human red blood cells (RBCs), and clinical ascites fluid.
Key findings include:
- Resolution Enhancement: In simulations, OmniFHT achieved a 1.75x improvement in resolution compared to standard Rytov-based methods, as measured by Fourier Shell Correlation (FSC).
- Robustness to Complex Motion: The system successfully resolved the biconcave shape of RBCs (1.42 $\mu$m resolution) and maintained accuracy even during cell-to-cell collisions, which typically render data unusable in traditional FHT.
- Sparsity and Efficiency: High-fidelity reconstructions were achieved with as few as 10 views or 120° of angular coverage, significantly reducing the data requirements for 3D imaging.
- Clinical Application: The authors demonstrated the first in situ 3D reconstruction of an entire clinical population from an ascites specimen, enabling the identification of diverse cell types (lymphocytes, neutrophils, and cancer cells) without manual pre-selection.
Discussion and Future Directions
OmniFHT represents a significant leap toward unbiased, high-throughput 3D flow cytometry. By eliminating the need for controlled cell rotation, it allows for the analysis of heterogeneous populations, including rare and irregularly shaped cells. Future work will focus on increasing computational speed through hardware acceleration (GPUs/FPGAs) and incorporating biological priors to further enhance robustness in high-noise environments or near-wall flow conditions.
Summary updated