The performance of an optical imaging system is typically characterized by the intensity Point Spread Function(PSF)or Optical Transfer Function(OTF)[
Acta Photonica Sinica, Volume. 52, Issue 9, 0911001(2023)
Quantitative Retrieval of Coherent Transfer Function via Fourier Ptychographic Microscopy
The performance of an optical imaging system is typically characterized by the intensity Point Spread Function (PSF) or Optical Transfer Function (OTF). But the Coherent Transfer Function (CTF) is better for describing the coherent optical imaging system. Though the CTF characterizes the complex amplitude transfer properties of the light field, it is hard to measure compared with PSF. Fourier Ptychographic Microscopy (FPM) is a promising computational technique that can obtain both complex amplitude information of an object and the CTF of coherent imaging system, which provides a way to retrieve the CTF. FPM, combining the concept of aperture synthesis and phase retrieval, is a recently developed imaging technique that allows the reconstruction of high-resolution complex images with an extended field of view. By acquiring a series of low-resolution brightfield and darkfield images under inclined illumination and stitching them together in the Fourier domain, FPM can break through the frequency limit of the employed objective determined by its numerical aperture. Consequently, the space-bandwidth product of the optical imaging system can be effectively increased without precise mechanical scanning. The flexibility with low-cost hardware requirements makes FPM a powerful tool particularly potential for imaging biomedical samples in the field of digital pathology. Although many advanced FPM techniques have been proposed to achieve higher data acquisition efficiency and recovery accuracy in the past few years, little is known about the precision, stability, and requirements of the CTF, especially when there are inevitable system errors. If FPM can retrieve high-precision CTF, it will provide a new means for CTF calibration. Therefore, this thesis mainly studies the acquisition of CTF with high precision, stability and efficiency via FPM. In this paper, we investigate the reconstruction quality of the CTF under different system errors with different targeted algorithms and find that the reconstructions of CTF is more robust than the reconstructions of object. In addition, under the condition of good recovery of object function, different objective algorithms can also recover basically the same CTF. Therefore, the CTF recovered by FPM algorithm can be used to quantitatively characterize coherent optical systems. Based on this, we report a sub-region translation method named ST-FPM, which is used in Fourier ptychographic microscopy imaging. Based on the basic assumption that the aberration of adjacent local fields is basically unchanged, asymmetric spatial information is introduced to eliminate the grid noise caused by periodic illumination, which improves the recovery accuracy of CTF and accelerates the convergence speed of CTF reconstruction in limited images. The recovered CTF is deconvolved with incoherent images. And the contrast is additionally improved compared with the traditional FPM. In addition, this method can realize image refocusing without the prior information of defocus. In addition, we study the spatial and frequency domain data redundancy of Fourier ptychographic microscopy to recover the coherent transfer function, and find that at least about 40% spectral overlap rate is needed to accurately reconstruct the coherent transfer function, which is 10% higher than that without aberration. And at least 25 original low-resolution images are needed for the stability of coherent transfer function. Finally, we discuss the necessary conditions for stable CTF reconstruction, and verify the conclusion in simulation and experiment.
0 Introduction
The performance of an optical imaging system is typically characterized by the intensity Point Spread Function(PSF)or Optical Transfer Function(OTF)[
Fourier Ptychographic Microscopy(FPM)[
However,the performance of FPM reconstruction critically depends on the quality of the original data. The reconstruction suffers a lot from the artifacts caused by various kinds of systematic errors. Recently,a series of correction and optimization algorithms have been invented and complemented to deal with the issue. OU X et al. presented Embedded Pupil Function Recovery Fourier Ptychographic Microscopy(EPRY-FPM)algorithm[
In this article,we proved the robustness of CTF based on error correction simulations,which provides a theoretical foundation for quantitative calculation. Based on this conclusion,we proposed a sub-region translation strategy of FPM,termed Sub-region Translation Fourier Ptychographic Microscopy(ST-FPM)method,which allows higher-quality pupil reconstruction within a finite number of iterations even when the number of captured images is limited. It has been proved that the method is well suited for grid noise removal generally existing in the reconstruction of transparent samples. And it can realize image refocusing without adjusting the aberration correction term when the defocus distance is changed. Additionally,images deconvoluted by the reconstructed pupil functions present a significant improvement of contrast compared with the originally collected incoherent images because ST-FPM obtains extra redundant information from adjacent sub-fields when reconstructing pupil function. In addition,we explored the spatial and spectrum data redundancy of FPM to retrieve CTF and found at least approximately 40% overlap rate is required for accuracy reconstructions of CTF,which is 10% higher than aberration-free condition. And 25 raw images are required at least for the stability of CTF. Finally,we discussed the conditions required for stable CTF reconstruction and validated the conclusions in simulations and experiments.
1 Methods
1.1 FPM principles
The LED array of a typical FPM system is utilized for multiple-angle illumination. Assuming that a sufficiently thin sample with transmittance
Information contained in images acquired under different illumination angles can be synthesized through an iterative phase retrieval process to generate high-resolution object images including both amplitude and phase properties.
We built an experimental FPM platform composed of a programmable 32×32 LED array with 4 mm adjacent distance,a 4×/0.1 NA objective lens,and a 16-bits digital camera(1 280×960 pixels,3.75 μm). All high-resolution complex images are reconstructed using the sequential Gauss-Newton algorithm [
1.2 Influence of systematic error on CTF
Although FPM shows great prospects for high-resolution image reconstruction with large FOV,some systematic errors will inevitably cause artifacts in images,posing a great challenge to the quality of reconstruction.
Figure 1.Simulation results of FPM reconstruction with several systematic errors added in
where f(x,y)is reconstructed image,g(x,y)is referenced image,X×Y is the pixels of image. The closer RMSE is to 0,the closer the reconstruction quality of the image is to the true value,and the better the reconstruction quality is[
where x and y are two virtual nonnegative images,which have been aligned with each other,
It can be seen that both the reconstructed images and pupil functions are contaminated with irregular artifacts to varying degrees compared with ground truth. Existing optimization methods are generally designed for improving the reconstruction quality of object functions. We wonder whether the process will produce undesirable influence on pupil reconstruction. Therefore,we then utilize multiple optimization and correction methods to deal with these systematic errors. As shown in
Figure 2.Simulation results of systematic error correction with multiple algorithms
1.3 Sub-region translation FPM
Periodic LED illumination often results in grid noise in FPM reconstruction,especially for transparent samples[
As shown in
Figure 3.Diagram of sub-region translation FPM for pupil reconstruction
It should be emphasized that block processing is almost a compulsory option to be performed. On the one hand,LED illumination strictly belongs to spherical wave illumination due to its large divergence. When the sample is not placed far away from the light source,the illumination cannot be approximated to plane wave illumination. However,the approximation is established for a sub-region after block processing,thus the limitation of illumination distance can be relaxed. On the other hand,hundreds of captured images produce large data amounts and calculation burdens. Block processing helps to improve computation efficiency and can be used for parallel operation. The influence of coherence and vignetting effect is also non-negligible.
2 Results
When imaging transparent samples,phase information can be obtained quantitatively through traditional FPM. However,the periodic LED array used in lighting will cause obvious grid noise artifacts in phase image and pipul,which greatly affects the quantitative phase acquisition ability of FPM. At present,an existing solution is to use a non-uniform LED illumination array to solve the grid noise artifact by deviating the translational symmetry of sampling. However,this scheme needs to design a special LED array,which is expensive and not universal. In this part,ST-FPM algorithm is used to eliminate grid noise artifacts. In order to examine the grid noise removal performance of ST-FPM method,we compared the reconstruction results of a transparent Sigma Star sample using the traditional method and ST-FPM method in
Figure 4.Comparison of grid noise removal from sigma star transparent sample
Different from coma or astigmatism,defocus is very unique among all kinds of aberrations because it is not caused by inherent physical defects in optical imaging systems. In a“perfect”optical system,defocus aberration can still appear as long as the target sample deviates from the original focal plane of the optical system. At present,an existing method is to“digitally refocus”the image by numerically zeroing the defocus aberration[
Here,we proved that ST-FPM can realize image refocusing without the tedious adjustment. To examine the refocus ability of ST-FPM method,We compared the reconstruction results of a label-free U2OS sample using the traditional method and ST-FPM method in
Figure 5.Comparison of focusing from U2OS sample
Another intriguing application of ST-FPM method is to improve the contrast of images after performing a deconvolution operation on the obtained pupil function. We collected the images of the USAF target for pupil reconstruction using the illumination of central 5×5 LEDs.
Figure 6.Comparison of image contrast improvement after deconvolution using reconstructed pupil functions
Figure 7.Simulation results of the influence of spatial sampling rate and spectral overlapping rate on CTF reconstruction
Figure 8.Simulation results of the influence of captured image number and iteration number on CTF reconstruction
Figure 9.Experimental results of the influence of captured image number and iteration number on CTF reconstruction
Figure 10.Pupil images reconstructed by three schemes of ST-FPM with 5%~50% pixels translation distances in simulations
where
Both pupil functions reconstructed by traditional method and ST-FPM method enhance the contrast of images after deconvolution,with a larger improvement by ST-FPM. Traditional FPM obtains redundant information by synthesizing NA,which improves the contrast of the image. On this basis,ST-FPM can obtain extra redundant information from adjacent sub-fields when reconstructing pupil function,which results in higher contrast.
3 Discussions
3.1 Conditions required to reconstruct a stable CTF
FPM reconstruction requires a series of low-resolution intensity images whose corresponding sub-apertures are overlapping in the frequency domain. Spectral overlapping rate is defined by the ratio of the overlapping area of two adjacent CTFs to the area of a single aperture and can be calculated by
where
where λ is the wavelength of the illumination,M is the magnification of objective,and Δx is the pixel size of the camera. Higher spectral overlapping rates provide better CTF reconstruction quality,and increasing the spatial sampling rate can reduce the spectral overlapping rate required for stable CTF reconstruction. Overall,reconstruct a stable CTF requires at least approximately 40% spectral overlapping rate.
The captured data amount and convergence performance of reconstruction algorithms also have influence on reconstruction quality.
In the experiment,central 17×17 LEDs of the array provide an illumination of 518 nm wavelength. Since the interval between adjacent LEDs is fixed,the illumination height should be adjusted to obtain different values of spectral overlapping rate.
3.2 Preferences of ST-FPM method
We originally designed three translation schemes for ST-FPM method:2-direction translation,4-direction translation,and 8-direction translation. With more translation directions,the time required for reconstruction is longer. The order of different directions has almost no effect on the result. We choose to uniformly select the translation order in Fig3. We simulated to determine the optimal choice of translation direction and distance so that a good balance can be achieved between reconstruction quality and cost of time. For each translation scheme,we set the translation distance ranging from 5%~50% pixels of the sub-region size. The central 5×5 LEDs are utilized for data acquisition and the iterative reconstruction algorithm runs for a fixed number of 120 times.
4 Conclusion
In this work,the influence of systematic errors on FPM pupil recovery is investigated. Although currently available correction and optimization algorithms targeted at object reconstruction do not directly involve the update of pupil function,the quality of reconstructed pupil is still improved. Therefore,pupil function is robust during iterations and can be used for quantitative calculation. Based on this conclusion,we proposed a sub-region translation strategy of FPM,termed ST-FPM method,which allows higher-quality pupil reconstruction within a finite number of iterations even when the number of captured images is limited. It has been proved that the method is well suited for grid noise removal generally existing in the reconstruction of transparent samples. And it can focus the defocused image without adjusting the aberration correction term every time the defocus distance is changed. Additionally,images deconvoluted by the reconstructed pupil functions present a significant improvement of contrast compared with the originally collected incoherent images. Compared with traditional FPM,ST-FPM can obtain extra redundant information from adjacent sub-fields when reconstructing pupil function,which results in higher contrast.
We also studied the conditions required for stable CTF reconstruction in FPM. In terms of information redundancy,at least approximately 40% spectral overlapping rate achieves desirable reconstruction results without oversampling. And 25 raw images are required at least for the stability of CTF. Three factors,including spectral overlapping rate,the number of captured images,and iteration number,produce mutual trade-offs for CTF reconstruction. CTF can be better reconstructed with more images collected when the iteration number is fixed. Increasing the iteration number compensates for the shortcoming caused by using fewer images. When the spectral overlapping rate is higher,less information redundancy can be obtained,thus larger number of images and iterations are required to reconstruct CTF stably.
It should be noted that we obtained a higher contrast image by deconvolution of pupil function,which only indirectly proved that ST-FPM quantitatively recovered the correct CTF. The CTF still needs to be measured directly by the traditional wavefront detection method for comparison.
[2] J W GOODMAN. Introduction to fourier optics (4th Ed)(2017).
[9] A PAN, Y GAO, A WANG et al. High-throughput full-color Fourier ptychographic microscopy for the next generation of digital pathologic imager and analyser (invited). Acta Photonica Sinica, 51, 0751408(2022).
[30] Y GUO, L WU, S JING. A phase-truncated fourier transform asymmetric optical compression and encryption system. Acta Photonica Sinica, 51, 0610001(2022).
[35] H Y WU, M Q WANG, X LU et al. Fourier Ptychographic microscopy color imaging with simulated-annealing chromatic aberration correction. Acta Photonica Sinica, 50, 0311002(2021).
Get Citation
Copy Citation Text
Jiaming CHEN, An PAN, Aiye WANG, Caiwen MA, Baoli YAO. Quantitative Retrieval of Coherent Transfer Function via Fourier Ptychographic Microscopy[J]. Acta Photonica Sinica, 2023, 52(9): 0911001
Category:
Received: Apr. 1, 2023
Accepted: May. 10, 2023
Published Online: Oct. 24, 2023
The Author Email: PAN An (panan@opt.cn)