Laser & Optoelectronics Progress, Volume. 62, Issue 15, 1511001(2025)
CUDA-Accelerated Fourier Ptychographic Microscopy (Invited)
Fourier ptychographic microscopy (FPM) enables high-resolution imaging over a wide field of view; however, its reconstruction process is time-consuming. While deep-learning-based methods can substantially accelerate FPM reconstruction, they demonstrate poor generalization capability. To address this limitation, this study developed cuFPM, a CUDA-based method designed to expedite FPM reconstruction. Building on the feature-domain FPM method, a loss function incorporating edge features and its analytical gradient with respect to the optimization parameters were derived. Subsequently, parallelizable parts of the gradient computation process were identified, and parallel large-scale FPM gradient calculations were implemented using CUDA. Mini-batch stochastic gradient descent and the RMSProp optimizer were used to optimize both the complex amplitude of the reconstructed sample and the optical transfer function of the imaging system. Based on NVIDIA RTX 3090, with a reconstruction upsampling ratio of 8, cuFPM achieved average reconstruction time of 2.1 s for subregions of 512 × 512 × 145 FPM data and 43 s for those of 2048 × 2048 × 145 FPM data. The reconstructed field of view was 11 mm2, with a spatial resolution of 1024 lp/mm and a spatial bandwidth product of 165 Mpixel. cuFPM was approximately 30 times faster than CPU-based methods and six times faster than gpuArray-based methods implemented in MATLAB. On a simulation-based dataset, cuFPM achieved an average peak signal-to-noise ratio of 29.2 dB and a structural similarity index of 0.91. These results demonstrate that cuFPM substantially accelerates FPM reconstruction, providing a solution for advancing the engineering deployment of FPM and stacked imaging systems.
Get Citation
Copy Citation Text
Shuhe Zhang, Liangcai Cao. CUDA-Accelerated Fourier Ptychographic Microscopy (Invited)[J]. Laser & Optoelectronics Progress, 2025, 62(15): 1511001
Category: Imaging Systems
Received: Mar. 19, 2025
Accepted: Apr. 7, 2025
Published Online: Aug. 11, 2025
The Author Email: Liangcai Cao (clc@tsinghua.edu.cn)
CSTR:32186.14.LOP250850