CN-121998862-A - Artifact evaluation method for three-dimensional super-resolution microscopic imaging

CN121998862ACN 121998862 ACN121998862 ACN 121998862ACN-121998862-A

Abstract

The invention discloses an artifact evaluation method for three-dimensional super-resolution microscopic imaging, and belongs to the field of fluorescence microscopic imaging. The invention scales the high-resolution image to obtain the optimal three-dimensional resolution scaled image, then carries out linear transformation to obtain the three-dimensional resolution scaled image after the linear transformation, analyzes the three-dimensional resolution scaled image and the low-resolution image to solve and output an artifact evaluation index, judges the authenticity of the high-resolution image by analyzing the interpretable degree of an optical imaging model between the low-resolution image before super resolution and the high-resolution image after super resolution, and quantitatively evaluates the artifact condition in the three-dimensional super resolution technology by a calculation method with lower cost. The method is suitable for evaluating the super-resolution artifact conditions of three-dimensional super-resolution microscopes such as a three-dimensional structured light illumination microscope, a three-dimensional single-molecule positioning super-resolution microscope and the like, and can also be used for evaluating the super-resolution artifact conditions of three-dimensional deconvolution and three-dimensional deep learning networks.

Inventors

XI PENG
HOU YIWEI
LI MEIQI

Assignees

北京大学

Dates

Publication Date: 20260508
Application Date: 20260115

Claims (10)

1. An artifact evaluation method for three-dimensional super-resolution microscopic imaging, which is characterized by comprising the following steps: 1) Acquiring an evaluation image pair: obtaining a low-resolution image and a high-resolution image, wherein the two images are paired images of the same visual field; 2) Obtaining a three-dimensional resolution scaled image: a) Obtaining a three-dimensional resolution scaling function based on Gao Siluo Lorentz functions, wherein the three-dimensional resolution scaling function is related to transverse and axial three-dimensional resolution scaling parameters; b) Defining a three-dimensional resolution scaling image of the high-resolution image according to the three-dimensional resolution scaling function; c) Setting a loss function based on the Pearson correlation coefficient, and carrying out parameter search by adopting a genetic algorithm to obtain an optimal three-dimensional resolution scaling parameter of the three-dimensional resolution scaling image; d) According to the obtained optimal three-dimensional resolution scaling parameters of the three-dimensional resolution scaling image, scaling the high-resolution image according to the definition of the three-dimensional resolution scaling image in the step b), so as to obtain the optimal three-dimensional resolution scaling image; 3) Linear transformation: Defining linear transformation of the optimal three-dimensional resolution scaled image, setting a loss function based on an average absolute error, carrying out parameter search by adopting a genetic algorithm to obtain a slope and a bias coefficient of the optimal linear transformation, and carrying out linear transformation on the optimal three-dimensional resolution scaled image to obtain a three-dimensional resolution scaled image after linear transformation; 4) Artifact evaluation: And analyzing the three-dimensional resolution scaled image and the low-resolution image after linear transformation, and solving and outputting an artifact evaluation index.
2. The artifact evaluation method according to claim 1, wherein in step 1), a three-dimensional super-resolution microscope is used to collect a low-resolution image and a high-resolution image of the same field of view, respectively, or a low-resolution microscope is used to collect a low-resolution image, and a three-dimensional super-resolution calculation process is performed on the low-resolution image to obtain a high-resolution image of the same field of view, wherein the number of pixels of the high-resolution image is identical to that of pixels of the low-resolution image.
3. The artifact estimation method according to claim 1, wherein in step 2) a) the three-dimensional resolution scaling function RSF based on Gao Siluo lorentz functions is defined as: Where σ xy and σ z are the lateral and axial three-dimensional resolution scaling parameters, respectively, and x, y and z are the three-dimensional coordinates of the image.
4. A method of assessing an artifact according to claim 3 wherein in step 2) b), the high resolution image is convolved with a three-dimensional resolution scaling function, the resolution of the high resolution image being scaled to obtain a three-dimensional resolution scaled image, the three-dimensional resolution scaled image I Sσ being defined as: Wherein I S is a high resolution image.
5. The artifact estimation method according to claim 4, wherein in step c) of 2), the three-dimensional resolution scaling parameter search is defined as follows using a genetic algorithm based on pearson correlation coefficient setting loss function: Wherein I D is a low-resolution image, the PCC represents a pearson correlation coefficient operation, and the pearson correlation coefficient operation PCC is defined as follows: Where N is the number of pixels, I Di is the I-th pixel of the low-resolution image, and I Sσi is the I-th pixel in the three-dimensional resolution scaled image.
6. The artifact estimation method according to claim 5, wherein the number of search iterations of the genetic algorithm is set to 40-60, and the search value ranges of the transverse and axial three-dimensional resolution scaling parameters σ xy and σ z are each 0-5.
7. The artifact estimation method according to claim 5, wherein in step d) of step 2), the best three-dimensional resolution scaled image I S scaled is: Where σ xy 'is σ z ' is the lateral and axial best three-dimensional resolution scaling parameters, respectively.
8. The artifact estimation method according to claim 7, wherein the linear transformation coefficient parameter search using a genetic algorithm is defined as follows, based on the mean absolute error set-up loss function: where a and b are the slope and bias coefficients of the linear transformation, respectively.
9. The artifact estimation method of claim 8, wherein the linear transformation is performed on the optimal three-dimensional resolution scaled image, and the three-dimensional resolution scaled image I S affine after the linear transformation is: Wherein, the And The slope and bias coefficients of the optimal linear transformation, respectively.
10. The artifact estimation method according to claim 9, wherein in step 4), the artifact estimation index comprises a three-dimensional resolution scaled pearson correlation coefficient 3D-RSP, an artifact map ErrorMap, and a three-dimensional resolution scaled error 3D-RSE: wherein ErrorMap i is the ith pixel of the artifact map.

Description

Artifact evaluation method for three-dimensional super-resolution microscopic imaging Technical Field The invention relates to the field of fluorescence microscopy imaging, in particular to an artifact evaluation method of three-dimensional super-resolution microscopy imaging. Background Fluorescence microscopes such as wide-field and confocal are important tools for life science observation and research. However, the resolution is limited by the fluctuation of light. Meanwhile, as the fluorescence microscope adopts an object lens to collect the emitted light for imaging, the axial resolution is usually only one third of the lateral resolution when three-dimensional imaging is carried out. To overcome the three-dimensional resolution limitation, methods by physics and computation have been proposed. For example, three-dimensional structured light illumination microscopy (3D-SIM) can realize frequency spectrum movement and reconstruction of three-dimensional high-frequency information through three-beam interference, and the three-dimensional resolution of a wide-field microscope is improved by two times. In the calculation method, the three-dimensional deconvolution method can improve the resolution of three-dimensional fluorescence imaging based on a physical model. The deep learning method can learn the simulated degradation process or acquire the physical super-resolution result to train the network to improve the resolution of the diffraction-limited three-dimensional image. However, these three-dimensional super-resolution methods can be subject to artifacts and errors due to imperfections in their implementation. For example, the 3D-SIM suffers from noise and super-resolution reconstruction artifacts may occur. Errors and artifacts can also occur with the three-dimensional deconvolution method due to deviations of the model assumptions from the real world situation. The limited generalization capability of the depth network can also lead to errors and artifacts in the actual use. Document [1] presents a method of evaluating two-dimensional planar super-resolution artifacts, but studies also indicate that this method cannot be used to evaluate three-dimensional images. There is still a lack of suitable artifact tools for assessing three-dimensional imaging. Disclosure of Invention Aiming at the problems existing in the prior art, the invention provides an artifact evaluation method for three-dimensional super-resolution microscopic imaging, which is used for quantitatively evaluating artifacts existing in the three-dimensional super-resolution technology. The invention relates to an artifact evaluation method of three-dimensional super-resolution microscopic imaging, which comprises the following steps: 1) Acquiring an evaluation image pair: obtaining a low-resolution image and a high-resolution image, wherein the two images are paired images with the same visual field, and the low-resolution image is marked as I D, and the high-resolution image is marked as I S; 2) Obtaining a three-dimensional resolution scaled image: a) Obtaining a three-dimensional resolution scaling function based on Gao Siluo Lorentz functions, wherein the three-dimensional resolution scaling function is related to transverse and axial three-dimensional resolution scaling parameters; b) Defining a three-dimensional resolution scaling image of the high-resolution image according to a three-dimensional resolution scaling function, wherein the three-dimensional resolution scaling image is related to transverse and axial three-dimensional resolution scaling parameters; c) Setting a loss function based on the Pearson correlation coefficient, and carrying out parameter search by adopting a genetic algorithm to obtain an optimal three-dimensional resolution scaling parameter of the three-dimensional resolution scaling image; d) According to the obtained optimal three-dimensional resolution scaling parameters of the three-dimensional resolution scaling image, scaling the high-resolution image according to the definition of the three-dimensional resolution scaling image in the step b), so as to obtain the optimal three-dimensional resolution scaling image; 3) Linear transformation: Defining linear transformation of the optimal three-dimensional resolution scaled image, setting a loss function based on an average absolute error, carrying out parameter search by adopting a genetic algorithm to obtain a slope and a bias coefficient of the optimal linear transformation, and carrying out linear transformation on the optimal three-dimensional resolution scaled image to obtain a three-dimensional resolution scaled image Is affine after linear transformation; 4) Artifact evaluation: And analyzing the three-dimensional resolution scaled image and the low-resolution image after linear transformation, and solving and outputting an artifact evaluation index. In the step 1), a three-dimensional super-resolution microscope is adopted to respectively ac