CN-121998827-A - Single-molecule positioning super-resolution imaging method, device, equipment and medium

CN121998827ACN 121998827 ACN121998827 ACN 121998827ACN-121998827-A

Abstract

The application discloses a single-molecule positioning super-resolution imaging method, device, equipment and medium, which relate to the field of optical microscopy imaging and comprise the steps of acquiring a currently acquired single-molecule image containing a plurality of PSF light spots, inputting the single-molecule image into a single-molecule positioning model based on a depth residual error network, a characteristic pyramid network, a region generating network, roIAlign layers, a mask head and a key head, carrying out sub-pixel level positioning and brightness estimation on the PSF light spots in the current single-molecule image to obtain a plurality of single-molecule positioning points and positioning point brightness values, carrying out uncertainty prediction on the single-molecule positioning points and positioning point brightness values, screening the single-molecule positioning points based on positioning point uncertainty and brightness uncertainty, and superposing and rendering the screened positioning points to carry out super-resolution reconstruction on the single-molecule image so as to obtain a single-molecule super-resolution image. The application can improve the precision and efficiency of single-molecule positioning super-resolution imaging.

Inventors

ZHANG SHENHE
CHEN HONGBO
WANG DAPENG

Assignees

中国科学院长春应用化学研究所

Dates

Publication Date: 20260508
Application Date: 20260211

Claims (10)

1. The single-molecule positioning super-resolution imaging method is characterized by being applied to a single-molecule positioning super-resolution microscopic imaging system and comprising the following steps of: Acquiring a currently acquired single-molecule image containing a plurality of PSF light spots, and obtaining a current single-molecule image; Inputting the current single-molecule image into a trained single-molecule positioning model to perform sub-pixel level positioning and brightness estimation on PSF light spots in the current single-molecule image to obtain a plurality of single-molecule positioning points and positioning point brightness values, and performing uncertainty prediction on the single-molecule positioning points and the positioning point brightness values to obtain positioning point uncertainty and brightness uncertainty, wherein the single-molecule positioning model is a model obtained by training an initial model based on a depth residual error network, a characteristic pyramid network, a region generation network, roIAlign layers, a mask head and a key head by using a historical single-molecule image; Screening the single-molecule positioning points based on the positioning point uncertainty and the brightness uncertainty to obtain screened positioning points; and superposing and rendering all the screened positioning points to perform super-resolution reconstruction on the current single-molecule image so as to obtain a single-molecule super-resolution image.
2. The method of single-molecule positioning super-resolution imaging according to claim 1, wherein said performing sub-pixel level positioning and luminance estimation on the PSF light spot in the current single-molecule image to obtain a plurality of single-molecule positioning points and positioning point luminance values, and performing uncertainty prediction on the single-molecule positioning points and the positioning point luminance values to obtain positioning point uncertainty and luminance uncertainty, comprises: Performing multi-scale semantic feature extraction on PSF light spots in the current single-molecule image through the depth residual error network to obtain current single-molecule multi-scale features; Performing feature fusion on the current single-molecule multi-scale features through the feature pyramid network to obtain a current single-molecule feature map, and generating candidate areas on the current single-molecule feature map through the area generating network to obtain a plurality of single-molecule candidate frames; Carrying out pixel-level feature alignment on the features in the single-molecule candidate frames through the RoIAlign layers to obtain aligned candidate frames, and carrying out mask marking on the aligned candidate frames through the mask header to obtain marked candidate frames; Sub-pixel level positioning and brightness estimation are carried out on PSF light spots in the marked candidate frames through the key point head, single-molecule positioning points and positioning point brightness values corresponding to the marked candidate frames are obtained, uncertainty prediction is carried out on the single-molecule positioning points and the positioning point brightness values, and positioning point uncertainty and brightness uncertainty are obtained.
3. The method of single-molecule positioning super-resolution imaging according to claim 2, wherein said performing sub-pixel positioning and brightness estimation on the PSF light spot in each marked candidate frame by the key point head, to obtain a single-molecule positioning point and a positioning point brightness value corresponding to each marked candidate frame, comprises: Sampling the position features and the brightness features of the PSF light spots in the aligned candidate frames by using a bilinear interpolation method to obtain a plurality of sampled features corresponding to each aligned candidate frame, and encoding each sampled feature to obtain encoded features; Inputting the coded features corresponding to the aligned candidate frames to a preset linear mapping layer respectively to predict the position average value and the brightness average value of the central points of the aligned candidate frames; mapping the position mean value to the coordinate system of the current single-molecule image from the coordinate system corresponding to the aligned candidate frame to obtain single-molecule locating points corresponding to the aligned candidate frames, and taking the brightness mean value as the locating point brightness value of the single-molecule locating points.
4. The method of single-molecule localization super-resolution imaging as claimed in claim 3, wherein said performing uncertainty prediction on said single-molecule localization point and said localization point luminance value to obtain localization point uncertainty and luminance uncertainty comprises: Calculating the variance of the coded features to obtain an initial variance, and mapping the initial variance into a standard deviation in a logarithmic variance mapping mode to predict uncertainty of the single-molecule locating point and the locating point brightness value to obtain locating point uncertainty and brightness uncertainty.
5. The method of single-molecule positioning super-resolution imaging as claimed in claim 3, wherein said performing sub-pixel positioning and brightness estimation on the PSF light spot in each of the marked candidate frames by using the key point head, to obtain single-molecule positioning points and positioning point brightness values corresponding to each of the marked candidate frames, comprises: And carrying out sub-pixel level positioning and brightness estimation on PSF light spots in each marked candidate frame by using the key point head and a heat map method to obtain single-molecule positioning points and positioning point brightness values corresponding to each marked candidate frame.
6. The method of single-molecule localization super-resolution imaging according to claim 1, wherein the screening the single-molecule localization points based on the localization point uncertainty and the brightness uncertainty to obtain screened localization points comprises: removing a first positioning point with the positioning point uncertainty larger than a first credibility threshold value and a second positioning point with the brightness uncertainty larger than a second credibility threshold value from the single-molecule positioning points to obtain screened positioning points; Correspondingly, the overlapping and rendering are carried out on all the screened positioning points so as to carry out super-resolution reconstruction on the current single-molecule image, and a single-molecule super-resolution image is obtained, which comprises the following steps: superposing all the screened positioning points to obtain a current single-molecule positioning point cloud; And rendering the current single-molecule positioning point cloud by using a weighted kernel rendering technology so as to reconstruct the current single-molecule image in a super-resolution way, thereby obtaining a single-molecule super-resolution image.
7. The single molecule localized super-resolution imaging method of any one of claims 1 to 6, further comprising: Collecting a historical single-molecule image, and labeling the historical single-molecule image to obtain a single-molecule image after labeling; Inputting the marked single-molecule image into an initial model based on a depth residual error network, a characteristic pyramid network, a region generation network, roIAlign layers, a mask head and a key head, so as to train the initial model by utilizing the marked single-molecule image and adopting a negative log likelihood function based on a multi-element Gaussian, and obtaining the single-molecule positioning model.
8. The utility model provides a single molecule location super-resolution imaging device which characterized in that is applied to single molecule location super-resolution microscope, includes: the acquisition module is used for acquiring a currently acquired single-molecule image containing a plurality of PSF light spots to obtain a current single-molecule image; The input module is used for inputting the current single-molecule image into a trained single-molecule positioning model so as to conduct sub-pixel level positioning and brightness estimation on PSF light spots in the current single-molecule image to obtain a plurality of single-molecule positioning points and positioning point brightness values, and conducting uncertainty prediction on the single-molecule positioning points and the positioning point brightness values to obtain positioning point uncertainty and brightness uncertainty, wherein the single-molecule positioning model is a model obtained by training an initial model based on a depth residual error network, a characteristic pyramid network, a region generation network, roIAlign layers, a mask head and a key point head by utilizing a historical single-molecule image; The screening module is used for screening the single-molecule locating points based on the locating point uncertainty and the brightness uncertainty to obtain screened locating points; and the super-resolution imaging module is used for superposing and rendering all the screened positioning points so as to perform super-resolution reconstruction on the current single-molecule image and obtain a single-molecule super-resolution image.
9. An electronic device comprising a processor and a memory, wherein the processor implements the single molecule localized super-resolution imaging method of any one of claims 1 to 7 when executing a computer program stored in the memory.
10. A computer readable storage medium for storing a computer program, wherein the computer program when executed by a processor implements the single molecule localization super resolution imaging method of any one of claims 1 to 7.

Description

Single-molecule positioning super-resolution imaging method, device, equipment and medium Technical Field The application relates to the field of optical microscopic imaging, in particular to a single-molecule positioning super-resolution imaging method, device, equipment and medium. Background Single-molecule positioning microscope (Single-Molecule Localization Microscopy, SMLM) is a super-resolution fluorescence imaging microscopy technology, and can accurately position the nanometer level of a biological molecule by accurately tracking random flicker of Single fluorescent molecules (such as complex subcellular structures and dense protein clusters in cells) between a bright state and a dark state, the imaging resolution can reach tens of nanometers, and the diffraction limit of a traditional optical microscope is broken through, so that the Single-molecule positioning microscope is widely applied to the biomedical fields such as cell structure analysis, dynamic process observation and the like. The imaging principle of the single-molecule positioning super-resolution microscope is that by controlling fluorescent molecules to randomly flash, only part of fluorescent molecules in each frame are in a luminous state, the luminous state of the molecules is recorded by a camera, each molecule can form a PSF (point spread function ) light spot, a positioning point is obtained after sub-pixel positioning is carried out on the PSF light spot, and then all the positioning points in a multi-frame image are summarized and rendered, so that the super-resolution image is obtained. Currently, in order to improve the time resolution and the imaging speed, more fluorescent molecules in a single frame image are usually made to emit light simultaneously, so that the single frame molecular density is improved. However, under the high-density condition, PSF light spots of a plurality of molecules are seriously overlapped, and the traditional algorithm (such as single-molecule Gaussian fitting, multi-emitter fitting, compressed sensing and the like) is easy to generate missed detection, false detection and positioning deviation, is sensitive to factors such as initial values, threshold values, noise models, PSF deformation and the like, and needs to be manually and repeatedly adjusted, so that the imaging speed is reduced. In addition, under the condition of low Signal-to-Noise Ratio (SNR) or complex background, the positioning error can be further increased, and long-time acquisition can also introduce sample/platform drift (such as mechanical drift, thermal drift and the like) so as to lead the whole positioning point cloud to generate time-varying displacement, and if the positioning point cloud is not corrected, structural stretching, blurring and artifact can be caused, so that the reconstructed image is easy to generate phenomena such as oversharpening or structural fracture, thereby affecting the quantitative consistency and imaging quality of the reconstructed image, and the reconstructed image can not accurately reflect the real biological sample structure. In summary, how to quickly and accurately realize the positioning and reconstruction of single molecules under the comprehensive complex conditions of high-density overlapping PSF light spots, low SNR, long-term drift and the like, and obtaining super-resolution images is a problem to be further solved in the field at present. Disclosure of Invention In view of the above, the present application aims to provide a method, an apparatus, a device and a medium for single-molecule positioning super-resolution imaging, which can improve the precision and efficiency of single-molecule positioning super-resolution imaging. The specific scheme is as follows: in a first aspect, the application discloses a single-molecule positioning super-resolution imaging method, which is applied to a single-molecule positioning super-resolution microscope and comprises the following steps: Acquiring a currently acquired single-molecule image containing a plurality of PSF light spots, and obtaining a current single-molecule image; Inputting the current single-molecule image into a trained single-molecule positioning model to perform sub-pixel level positioning and brightness estimation on PSF light spots in the current single-molecule image to obtain a plurality of single-molecule positioning points and positioning point brightness values, and performing uncertainty prediction on the single-molecule positioning points and the positioning point brightness values to obtain positioning point uncertainty and brightness uncertainty, wherein the single-molecule positioning model is a model obtained by training an initial model based on a depth residual error network, a characteristic pyramid network, a region generation network, roIAlign layers, a mask head and a key head by using a historical single-molecule image; Screening the single-molecule positioning points based on the positioning point