A feature extraction method for cryo-electron tomography images

Abstract

The present invention discloses a feature extraction method for cryo-electron tomography images, including: S1 obtaining a projection image of a cryo-electron tomography and a corresponding image sequence; S2 presetting an angle sequence for affine transformation simulation; S3 affine transformation simulation of each projection image of a cryo-electron tomography according to the set angle sequence, and obtaining a corresponding simulated image set after screening; S4 extracting matching feature points according to the simulated image set obtained in S3 and the corresponding angle sequence. This method solves the interference of the spatial position of the sample during artificial tilting by simulating the atlas and the corresponding angle sequence, improves the extraction quality of feature points, and thus obtains a high-quality cryo-electron three-dimensional reconstruction model.

Description

Technical Field

[0001] The present invention relates to the technical field of structural biology, and in particular to a feature extraction method for cryo-electron tomography images. Background Art

[0002] Electron tomography three-dimensional reconstruction technology is an important branch of using cryo-electron microscopy to study the structure of biological macromolecules. It has irreplaceable advantages over other cryo-electron microscopy reconstruction technologies in the study of amorphous, asymmetric and non-uniform structures of biological samples.

[0003] The traditional electron tomography 3D reconstruction technology reversely reconstructs the 3D structure of a biological sample by acquiring 2D projection images at multiple continuous angles. During the projection imaging process, the direction of the number of electrons emitted by the transmission electron microscope remains fixed, and the sample is tilted along an axis perpendicular to the electron beam. The tilt range is approximately between ±60° and ±80°, with an interval of 1° to 2°. Each time the sample is tilted to a certain angle, a 2D projection image of the sample in the corresponding direction is taken, and then this series of 2D images is processed to obtain the 3D structure of the sample using a 3D reconstruction algorithm.

[0004] However, if the samples are artificially tilted during the shooting process, since most of the samples being photographed are at the nanometer level, the slight deviation in spatial movement during rotation will be infinitely magnified, thus affecting the imaging effect. Therefore, it is necessary to align the two-dimensional projection images at different angles before performing three-dimensional reconstruction, and correct the image sequence to align them in space.

[0005] Patent document CN 112614170A discloses a cryo-electron microscopy particle image registration method based on Fourier power spectrum. The method estimates the parameters of the power spectrum high-pass filter of the cryo-electron microscopy particle image set to be tested, calculates the Fourier power spectrum of the image to be registered, extracts feature points, and corrects the image after reverse rotation based on the feature points. This method effectively separates the noise, particles, and background of the image through the characteristics of the Fourier power spectrum, thereby improving the accuracy of image calibration. However, this method requires precise annotation of the original image, but does not take into account the influence of the spatial position deviation caused by rotation on the relative position of particles in the image.

[0006] Patent document CN113160152A discloses a cryo-electron microscopy single particle selection method based on image fusion and threshold segmentation, first input cryo-electron microscopy image data, and normalize the image data, then reduce the normalized image files in sequence, generate a series of images of different sizes, then restore the images of different scales to images of the same size as the normalized image, accumulate and average the images restored by the multi-scale images, and obtain the fused image; separate the particles and the background of the fused image, remove the abnormal area for the particle area image obtained by threshold segmentation, and finally calculate the contour center based on the particle area that completes the removal process, and obtain the seat mark of the particles in the cryo-electron microscopy image. This method can improve the efficiency of particle selection, but this method needs to screen each image and remove some particles at the same time, so that the number of particles in the entire image set is reduced, thereby affecting the accuracy of subsequent three-dimensional reconstruction. Summary of the invention

[0007] In order to solve the above problems, the present invention provides a feature extraction method for cryo-electron tomography images. The method obtains a set of simulated images corresponding to actual cryo-electron tomography images through affine transformation through a preset rotation angle sequence, and performs feature extraction based on clear rotation angles and corresponding simulated images, thereby avoiding errors caused by spatial movement when rotating the sample in actual operation, thereby increasing the number of effective feature points; further, the number of feature point alignment results is increased, thereby obtaining a high-quality three-dimensional reconstructed image.

[0008] A feature extraction method for cryo-electron tomography images, comprising:

[0009] S1 obtains the projection image of cryo-electron tomography and the corresponding image sequence;

[0010] S2 presets the angle sequence for affine transformation simulation;

[0011] S3 performs affine transformation simulation on each cryo-electron tomography projection image according to the set angle sequence, and obtains the corresponding simulation image set after screening;

[0012] S4 extracts matching feature points based on the simulated image set obtained in S3 and the corresponding angle sequence.

[0013] Preferably, the angle sequence of the preset affine change simulation in S2 is based on the rotation range of the electronic sample during the projection imaging in S1 from 0° to θ max and interval angle Δθ are set.

[0014] Preferably, the angle sequence is 0, Δθ, 2Δθ, 3Δθ ... kΔθ, where k = 0, 1, 2, 3 ... n,

[0015] Preferably, the S3 affine transformation simulation is based on the tilt parameters obtained by angle sequence transformation, and the simulation is completed using bilinear interpolation.

[0016] Preferably, the tilt parameter is specifically:

[0017]

[0018] Where t is the tilt parameter and θ is the tilt angle.

[0019] Based on the above calculation results, the transformation relationship between the image to be simulated and the simulated image is calculated:

[0020] (t0x,y)→u(t′x,y)

[0021] Where t0 is the tilt parameter of the image to be simulated, and t′ is the tilt parameter corresponding to the simulated image.

[0022] Preferably, the bilinear interpolation method is used to complete the simulation, specifically: resampling each projection image of the cryo-electron tomography along the rotation direction of the electronic sample to obtain a set of simulated images corresponding to the projection image of the cryo-electron tomography, wherein the resampling coordinate compensation value is

[0023] Preferably, the screening is based on comparing an image with the same tilt angle in the generated simulated image set with the corresponding actual electron tomography image, and when the pixel difference between the simulated image and the actual electron tomography image is less than a threshold, the simulated image is discarded.

[0024] Preferably, the threshold is set to 50-100 pixels, wherein the larger the pixel threshold is, the fewer images there are in the simulated image set and the higher the accuracy of the feature points; conversely, the more images there are in the simulated image set, the more feature points there are.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] (1) The projection images of actual cryo-electron tomography are processed by plane affine transformation to obtain a set of simulated images with a clear rotation angle, thus avoiding the spatial motion error caused by artificial rotation of the electron sample.

[0027] (2) By generating a set of simulated images and the corresponding rotation angles, the traditional SIFT feature map extraction method can be used to obtain more effective feature points, thereby providing more and more accurate matching alignment data for subsequent three-dimensional reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A flow chart of a feature extraction method for cryo-electron tomography images is provided for the present invention;

[0029] Figure 2 This is a comparison chart of the number of feature points of the data in this embodiment;

[0030] Figure 3 This is the reprojection error map after the feature points of the data in this embodiment are extracted using the traditional SIFT method;

[0031] Figure 4 This is a reprojection error map after feature points are extracted from the data of this embodiment using the method provided by the present invention. DETAILED DESCRIPTION

[0032] A set of experimental data was set, with the projection angle range from -60° to 60° and the interval between each projection being 3°. A total of 41 original cryo-electron tomography projection images were taken, each with a size of 5760×4092.

[0033] For this group of data, Figure 1 As shown, feature extraction of cryo-electron tomography images is performed, including:

[0034] S1 obtains the above experimental data, including its projection angle range, shooting interval and corresponding image sequence;

[0035] S2 is based on a projection angle range of -60° to 60°, a projection interval of 3°, and a preset angle sequence (0, 3°, 6°, ... 60°).

[0036] S3 performs a planar affine transformation on each cryo-electron tomogram in the image sequence based on a preset angle sequence (0, 3°, 6°, ... 60°):

[0037] First calculate the tilt parameter:

[0038]

[0039] Where t is the tilt parameter and θ is the tilt angle.

[0040] Based on the above calculation results, the transformation relationship between the image to be simulated and the simulated image is calculated:

[0041] (t0x,y)→u(t′x,y)

[0042] Where t0 is the tilt parameter of the image to be simulated, and t′ is the tilt parameter corresponding to the simulated image.

[0043] The projection image of each cryo-electron tomogram is resampled along the rotation direction of the electron sample by bilinear interpolation to obtain a set of simulated images corresponding to the projection image of the cryo-electron tomogram, where the resampling coordinate compensation value is

[0044] Finally, the images with the same tilt angle in the generated simulated image set are compared with the corresponding actual electron tomography image projection. When the pixel difference between the simulated image and the actual electron tomography image projection is less than 100 pixels, the simulated image is discarded.

[0045] S4 extracts feature points through the SIFT algorithm based on the simulated atlas and the corresponding angle sequence generated by the above method. The SIFT algorithm is a common method on the market, and its process will not be described in detail.

[0046] like Figure 2 As shown, the number of feature points obtained by the feature extraction method provided by the present invention is much greater than the number of feature points directly obtained by the traditional SIFT algorithm without any processing, which provides more matching feature points for subsequent three-dimensional reconstruction.

[0047] After completing the alignment of the projected image sequence, reprojection is performed according to the alignment parameters at the solution to obtain the projected position of each feature point, and the distance between the reprojected position and the original position is calculated.

[0048] like Figure 3 As shown in , it is the reprojection error map of feature point extraction by SIFT algorithm without any processing; Figure 4 As shown, it is a reprojection error map of feature point extraction by the feature extraction method provided by the present invention; wherein the dotted line is the position of the error average value. According to the comparison of the two error maps, it can be seen that the error distribution of the feature points extracted by the method provided by the present invention is smaller than that of the traditional SIFT algorithm, and the error average value is also smaller than that of the traditional SIFT algorithm. Therefore, it is explained that the method provided by the present invention has certain advantages over the traditional SIFT algorithm.

Claims

1. A feature extraction method for cryo-electron tomography images, characterized in that: include: S1 obtains the projection image of cryo-electron tomography and the corresponding image sequence; S2 presets the angle sequence for affine transformation simulation; S3 performs affine transformation simulation on each cryo-electron tomography projection image according to the angle sequence preset in S2, and obtains a corresponding simulation image set by screening; S4 extracts matching feature points based on the simulated image set obtained in S3 and the corresponding angle sequence.

2. The feature extraction method according to claim 1, characterized in that: The angle sequence of the preset affine change simulation in S2 is based on the rotation range of the electronic sample during the projection imaging in S1, which is 0° to θ max and interval angle Δθ are set.

3. The feature extraction method according to claim 2, characterized in that: The angle sequence is 0, Δθ, 2Δθ, 3Δθ ... kΔθ, where k = 0, 1, 2, 3 ... n, 4. The feature extraction method according to claim 1, characterized in that: The S3 affine transformation simulation is based on the tilt parameters obtained by angle sequence transformation, and the simulation is completed by using bilinear interpolation method.

5. The feature extraction method according to claim 4, characterized in that: The tilt parameters are specifically: Where t is the tilt parameter and θ is the tilt angle; Based on the above calculation results, the transformation relationship between the image to be simulated and the simulated image is calculated: (t0x,y)→u(t′x,y) Where t0 is the tilt parameter of the image to be simulated, and t′ is the tilt parameter corresponding to the simulated image.

6. The feature extraction method according to claim 4, characterized in that: The bilinear interpolation method is used to complete the simulation, specifically: resampling each projection image of the cryo-electron tomography along the rotation direction of the electron sample to obtain a simulated image set corresponding to the projection image of the cryo-electron tomography, wherein the resampling coordinate compensation value is 7. The feature extraction method according to claim 1, characterized in that: The screening is based on comparing an image with the same tilt angle in the generated simulated image set with the corresponding actual electron tomography image. When the pixel difference between the simulated image and the actual electron tomography image is less than a threshold, the simulated image is eliminated.

8. The feature extraction method according to claim 7, characterized in that: The threshold was set to 50-100 pixels.

Images