Holographic Imaging Optofluidic Platform for Particles’ Three-Dimensional Location and Reconstruction With Physics-Aware Strategy

Abstract

Digital holographic imaging is a label-free, non-invasive imaging technique that can capture three-dimensional locational and morphological information of cells/particles. Applying artificial intelligence to holographic reconstruction has significantly enhanced the speed and accuracy for applications such as cell/particle tracking and profiling. However, existing data-driven learning methods typically require massive amounts of training data with diverse environmental conditions. To achieve high fidelity, extracting three-dimensional information from intensity shot noise in the data remains a challenge. This study developed a physics-aware strategy that mitigates noise in three-dimensional location and reconstruction. The model was trained using a physics-aware strategy and images synthetically generated without the need for experiments or real-world samples. The performance of the strategy was validated using experimentally acquired low-resolution and noisy holograms of micro particles in an optofluidic chip. Results showed that high-quality three-dimensional location and reconstruction can be successfully extracted. It demonstrated comparable precision, but with higher speed, compared to conventional methods. The developed strategy improves its interpretability without access to experimental data during training. This solution can be used in an integrated holographic imaging optofluidic chip offering high throughput, which is particularly well-suited for point-of-care testing applications.