Fast Calculation of Computer Generated Holograms for 3D Photostimulation through Compressive-Sensing Gerchberg–Saxton Algorithm

Abstract

The use of spatial light modulators to project computer generated holograms is a common strategy for optogenetic stimulation of multiple structures of interest within a three-dimensional volume. A common requirement when addressing multiple targets sparsely distributed in three dimensions is the generation of a points cloud, focusing excitation light in multiple diffraction-limited locations throughout the sample. Calculation of this type of holograms is most commonly performed with either the high-speed, low-performance random superposition algorithm, or the low-speed, high performance Gerchberg–Saxton algorithm. This paper presents a variation of the Gerchberg–Saxton algorithm that, by only performing iterations on a subset of the data, according to compressive sensing principles, is rendered significantly faster while maintaining high quality outputs. The algorithm is presented in high-efficiency and high-uniformity variants. All source code for the method implementation is available as Supplementary Materials and as open-source software. The method was tested computationally against existing algorithms, and the results were confirmed experimentally on a custom setup for in-vivo multiphoton optogenetics. The results clearly show that the proposed method can achieve computational speed performances close to the random superposition algorithm, while retaining the high performance of the Gerchberg–Saxton algorithm, with a minimal hologram quality loss.

Real Time Generation of Three Dimensional Patterns for Multiphoton Stimulation

“…The results clearly show how the higher convergence speed of CS-WGS, makes it the ideal candidate for real-time applications. The GPU implementation of the algorithm proves, for real time applications, absolutely necessary, as similar spots patterns to those tested would require several seconds for computation with CS-WGS (Pozzi et al, 2019 ), and up to several minutes with WGS.…”

“…We have recently proved (Pozzi et al, 2019 ), how, on a CPU, a new algorithm (compressive sensing weighted Gerchberg-Saxton, CS-WGS), applying the principles of compressed sensing to the iterations of WGS can reduce its computational cost asymptotically close to the cost of RS, while maintaining the high quality of WGS holograms. Here, we present the implementation of CS-WGS on a low-cost consumer GPU, demonstrating that the algorithm is well-suited to GPU implementation, enabling video-rate computation of holograms with e > 0.9 and u > 0.9 for N < 100 and M < 1, 152 2 , ideally adaptable to feedback-based optogenetic control of neuronal networks.…”

Fabrication and optical manipulation of micro-robots for biomedical applications

“…However, the control of optical microrobot via HOT is more time-intensive, compared to time-sharing OT, which limits its real-time application. Therefore, the trade-off between performance and speed should be considered depending on different application requirements [18].…”

Optical Micromachines for Biological Studies

“…Several different methods for performing this calculation are presented in the literature, including the high-performance yet low-speed Gerchberg-Saxton algorithm [146] and the low-precision yet high-speed direct superposition method [147]. Therefore, when calculating holograms for optical tweezers or other laser-based applications, a trade-off must be made in terms of either speed or accuracy [148], with attempts made to adapt these existing algorithms to produce high speed, high performance alternatives [149,150]. Time-sharing set-ups for optical tweezers are based on rapid movement of the beam and typically use acousto-optic (AOD) or electo-optic deflectors (EOD), which deflect the beam by a certain angle, controlled by the frequency of an input signal [151,152].…”