Image adaptive point‐spread function estimation and deconvolution for in vivo confocal microscopy

Abstract

Visualizing deep inside the tissue of a thick biological sample often poses severe constraints on image conditions. Standard restoration techniques (denoising and deconvolution) can then be very useful, allowing one to increase the signal-to-noise ratio and the resolution of the images. In this paper, we consider the problem of obtaining a good determination of the point-spread function (PSF) of a confocal microscope, a prerequisite for applying deconvolution to three-dimensional image stacks acquired with this system. Because of scattering and optical distortion induced by the sample, the PSF has to be acquired anew for each experiment. To tackle this problem, we used a screening approach to estimate the PSF adaptively and automatically from the images. Small PSF-like structures were detected in the images, and a theoretical PSF model reshaped to match the geometric characteristics of these structures. We used numerical experiments to quantify the sensitivity of our detection method, and we demonstrated its usefulness by deconvolving images of the hearing organ acquired in vitro and in vivo.

Richardson-Lucy deconvolution for two-photon fluorescence images via non-linear diffusion equation pre-filtering

“…The PSF of two-photon laser scanning microscopy can be determined either experimentally or theoretically. However, the accurate PSF does not only depend on the optical paths of microscopy, but also depends on optical properties of specimen because it varies from sample to sample [4] [6] . In addition, noise is always present inevitably when we obtain measured PSF through experiment, while the theoretical PSF cannot account for subtle aberrations presented in optical system.…”

“…Some published literatures have demonstrated that no obviously distinction between the measured PSF and theoretical PSF when they are applied to image restoration, and the latter always has a little better result [4][5][6]] . So it is wise to employ the theoretical PSF as an initial one of the regularized R-L deconvolution algorithm [10] [12] .…”

“…However, the images are still suffered from blurring in neurobiological experiments due to random photon noise and optical system's distortion, especially imaging deep inside the thick specimens. In such cases, the photon noises induced by variations of light scattering and refractive index of specimen, result in a significant loss of fluorescence intensity and lower signal-to-noise ratio [4][5][6] . In order to improve the image quality and facilitate subsequent segmentation and three-dimension reconstruction, an effective and robust deconvolution scheme might be a good choice for restoring the original image besides fine-turning of optical paths of microscopy [5] .…”

Spatial quantitative analysis of fluorescently labeled nuclear structures: Problems, methods, pitfalls

“…In addition, PSFs may be different between samples and even within a single sample if the cells contain bodies with refractive indices that are significantly different from those of their surroundings (yolk granules, chromocenters, etc. ), as well as within thick samples (e.g., Holmes et al 2006;von Tiedemann et al 2006). Theoretically the use of blind deconvolution algorithms could alleviate these problems.…”

<title>Restoration of fluorescence images from two-photon microscopy using modified nonlinear anisotropic diffusion filter</title>

“…But most of deconvolution methods use the point spread function (PSF), they assume that the PSF of the two-photon laser scanning microscopy has been known or measured, or can be better estimated based on a priori knowledge. However, the PSF and prior knowledge of two-photon imaging system are always not available or inaccurate due to varied biologic specimens and specific experimental condition, and thus cannot bring a preferable restoration result [11] [12] . On the other side, the common denoising methods such as Bayesian estimate, maximum likelihood (ML) and maximum a posteriori (MAP) are used to removing image noise [13] .…”