CN113837947B - Processing method for obtaining optical coherence tomography large focal depth image - Google Patents

Abstract

The invention discloses a processing method for obtaining an optical coherence tomography large-focus-depth image, which mainly comprises three steps of constructing an OCT en face image data set, constructing an OCT en face image super-resolution model and realizing digital refocusing of an OCT image based on the constructed OCT en face image super-resolution model. In the construction of an image super-resolution model, the mapping relation between an OCT en face high-resolution image and a low-resolution image is learned by a depth learning method, the digital refocusing of an OCT en face three-dimensional image is obtained, and the expansion of the OCT image focal depth is realized. The digital refocusing mode is suitable for an OCT imaging mode, can quickly improve the transverse resolution at different depths in an OCT three-dimensional image, expands the focal depth, and can effectively reduce the hardware development difficulty of the OCT image refocusing.

Description

A processing method for obtaining large focal depth images of optical coherence tomography

technical field

The invention belongs to the technical field of image processing and imaging, and in particular relates to a processing method for obtaining a large focal depth image of optical coherence tomography.

Background technique

Optical coherence tomography (OCT) is a non-contact, non-invasive imaging technique. The lateral resolution of OCT is determined by the diffraction-limited spot size of the sample focused beam. As a 3D imaging technique, the higher the lateral resolution of OCT, the smaller the depth of focus. In order to obtain large depth imaging, a trade-off between lateral resolution and depth of focus is usually made in the development of OCT systems.

How to improve the depth of focus of OCT images in the case of high resolution has always been a difficult problem. A typical hardware approach is to use binary phase spatial filters or axial lenses to implement Bessel beams to extend the depth of focus. In addition, in 2017, E. Bo et al. used multiple aperture synthesis to expand the depth of focus (Bo E, Luo Y, Chen S, Liu X, Wang N, Ge X, WangX, Chen S, Chen S, Li J, Liu L. Depth-of-focus extension in optical coherencetomography via multiple aperture synthesis. Optica. 2017, 4(7), 701-706.). However, hardware methods usually require the development of complex imaging systems with poor stability.

Compared to hardware-based methods, digital signal processing methods offer alternative and relatively inexpensive solutions. For example, Interferometric synthetic aperture microscopy (ISAM) (Ralston T S, Marks D L, Carney P S, Boppart S A. Interferometric synthetic aperture microscopy. Nature physics. 2007; 3(2), 129-134.) can High-resolution imaging is performed in a large depth range, but absolute phase stability is required during acquisition.

SUMMARY OF THE INVENTION

The main purpose of the present invention is to solve the problem of how to obtain a large focal depth image while improving the lateral resolution, and to provide a processing method for obtaining a large focal depth image of optical coherence tomography.

The method of the invention first establishes a data set composed of OCT en face low-resolution images and high-resolution image pairs registered under different defocusing amounts, then constructs an OCT en face image super-resolution model, and learns the OCT en face high-resolution model through a deep learning method. According to the mapping relationship between the high-resolution image and the low-resolution image, the digital refocusing of the OCT three-dimensional image is obtained, so as to obtain the high-resolution image after the focal depth of the OCT image is expanded, that is, the large focal depth image.

To achieve the above purpose, the present invention adopts the following technical solutions:

A processing method for obtaining a large focal depth image of optical coherence tomography, comprising three steps: constructing an optical coherence tomography (OCT) en face image dataset, constructing an OCT en face image super-resolution model, and building an image based on the constructed OCT en face image. The OCT en face image super-resolution model realizes three steps of digital refocusing of OCT images;

Step 1: Construct the OCT en face image dataset;

The OCT en face image dataset refers to a dataset composed of pairs of OCT en face low-resolution images and high-resolution images registered under different defocus amounts; the specific construction method:

Step 1.1: Focus the incident beam on the inside of the sample, and collect the OCT en face image sequence at equal intervals for the selected area inside the sample; the specific method is as follows:

Move the sample stage along the specified direction, and collect the OCT en face image sequence of the selected area inside the sample when the incident beam is focused at M+1 positions, and the number of images collected in each image sequence is P; A set of OCT en face image sequences collected within the original focal depth range of the beam is used as a reference image sequence, and the reference image sequence belongs to the OCT en face high-resolution image; the rest are not within the original focal depth range of the incident beam. M positions The M groups of image sequences collected are OCT en face low-resolution image sequences under different defocus amounts, and the OCT en face low-resolution image sequences under different defocus amounts correspond to different resolutions;

The length of the selected area inside the sample is equal to 0.8-1 times the original focal depth of the incident beam;

The M is a natural number between 2-8;

The number P of images collected by each image sequence is a natural number between 4 and 100;

Step 1.2: For the OCT en face low-resolution image sequences with different resolutions, use the OCT en face image registration method in the reference image sequence to find and register the OCT en face high-resolution images at the corresponding positions, and establish M different distances. Data sets under the focal volume, each data set contains P registered OCT en face low-resolution image and high-resolution image pairs;

Wherein, the OCT en face image registration method includes two steps: the selection of the OCT en face low-resolution image and the high-resolution image pair, and the fine registration of the OCT en face low-resolution image and the high-resolution image;

(1) The selection method of the OCT en face low-resolution image and the high-resolution image pair is, for each defocused OCT en face extracted from the OCT en face low-resolution image sequence collected from a certain defocus position; A low-resolution image, select an OCT en face high-resolution image from the reference image sequence for pairing;

The specific method for obtaining an image pair: preliminarily register a certain defocused OCT en face low-resolution image with each OCT en face high-resolution image in the reference image sequence; then, calculate the defocus after the preliminary registration The correlation coefficient r between the OCT en face low-resolution image and each of the OCT en face high-resolution images, from which the OCT en face high-resolution image with the highest correlation coefficient and the out-of-focus OCT en face low-resolution image are selected as paired images ;

The correlation coefficient r between the two images is calculated by the following formula:

(1)

(1)

代表高分辨图像的平均灰度值，

代表低分辨图像的平均灰度值。where f(x,y) and g(x,y) represent the gray value of the high-resolution image and low-resolution image respectively, x represents the abscissa of the image, y represents the ordinate of the image,

represents the average gray value of the high-resolution image,

Represents the average gray value of the low-resolution image.

The method for the preliminary registration of the low-resolution image and the high-resolution image of the OCT en face includes an affine transformation method, a rigid body transformation method, a projection transformation method, a nonlinear transformation method or a moment and principal axis method;

(2) The fine registration methods of OCT en face low-resolution image and high-resolution image include pyramid registration method, wavelet transform registration method, maximum mutual information registration method and atlas registration method;

Step 2: Build the OCT en face image super-resolution model;

The deep learning method is used to learn the mapping relationship between the OCT en face high-resolution images and M sets of OCT en face low-resolution images with different resolutions, and construct M OCT en face image super-resolution models;

Step 3: realize digital refocusing of OCT images based on the OCT en face image super-resolution model;

Step 3.1: For the OCT 3D image data to be processed, first determine the focal position of the incident beam, divide the OCT 3D image to be processed according to the focal position, the original focal depth and the defocus amount, and obtain the OCT 3D image within the original focal depth range. The OCTen face high-resolution image sequence and each M group in the depth direction before the focal depth range and after the focal depth range, a total of 2M groups of OCTen face low-resolution image sequences; the OCTen face high-resolution image does not require super-resolution processing;

Step 3.2: Use the M OCT en face image super-resolution models constructed in Step 2 to carry out M groups of OCT en face low-resolution image sequences before and after the focal depth range divided under different defocusing amounts, respectively. Image super-resolution processing, obtaining each M group of OCT en face super-resolution image sequences before and after the focal depth range, realizing image refocusing, and improving the OCT en face image beyond the original focal depth in the OCT 3D image to be processed. Finally, a set of OCT en face high-resolution image sequences within the focal depth range without super-resolution processing and the processed 2M sets of OCT en face super-resolution image sequences are respectively taken along the depth direction. After the focal depth range is re-stacked to form a refocused OCT 3D image, the focal depth is expanded.

Beneficial effects of the present invention:

1. The present invention does not require the assistance of any mechanical hardware device, and realizes the expansion of the focal depth of the OCT image through a digital method, which can reduce the hardware cost of system development;

2. The present invention uses the deep learning method to realize the digital refocusing of the OCT image in combination with the image registration algorithm, and the processing speed is fast;

3. The implementation of the present invention has low requirements on the OCT hardware system, does not require phase matching, and has strong generalization ability.

Description of drawings

Fig. 1 is the processing method flow chart of obtaining optical coherence tomography large focal depth image provided by the present invention;

Fig. 2 is a schematic diagram of acquisition of OCT en face image sequence in a selected area inside the sample provided by the present invention; wherein (a) is a schematic diagram of the acquisition of a reference image sequence, and (b)-(e) are OCT en face under 4 different defocus amounts respectively. Schematic diagram of face low-resolution image sequence acquisition;

Fig. 3 is the OCT en face image pair registration method flow chart provided by the present invention;

Fig. 4 is the generator structure schematic diagram of the present invention;

5 is a schematic structural diagram of a residual dense connection block in the generator of the present invention;

Fig. 6 is the structure schematic diagram of the discriminator of the present invention;

Fig. 7 is the zebrafish OCT en face low-resolution image collected by the present invention;

Fig. 8 is the zebrafish OCT en face high-resolution image collected by the present invention;

Fig. 9 is a digital refocusing image of zebrafish OCT en face realized by the present invention.

Detailed ways

The realization, functional characteristics and advantages of the present invention will be further described below with reference to the accompanying drawings.

A processing method for obtaining a large focal depth image of optical coherence tomography, the flowchart of which is shown in accompanying drawing 1, including:

Step 1: Construct the OCT en face image dataset;

The OCT en face image dataset refers to a dataset composed of pairs of OCT en face low-resolution images and high-resolution images registered under different defocus amounts; the specific construction method:

The incident beam is focused inside the sample, and the OCTen face image sequence is collected at equal intervals for the selected area inside the sample; the specific method is as follows:

Move the sample stage along the specified direction, and collect the OCT en face image sequence of the selected area inside the sample when the incident beam is focused at M+1 positions, and the number of images collected in each image sequence is P; A set of OCT en face image sequences collected within the original focal depth range are taken as the reference image sequence, and the reference image sequence belongs to the OCT en face high-resolution images; the rest are collected from M positions that are not within the original focal depth range of the incident beam. The M groups of image sequences are OCT en face low-resolution image sequences under different defocus amounts, and the OCT en face low-resolution image sequences under different defocus amounts correspond to different resolutions; the OCT en face image registration method is used to find And register the high-resolution OCT en face images at the corresponding positions, and establish M data sets under different defocus amounts, each data set contains P registered OCT en face low-resolution images and high-resolution image pairs;

The specific implementation method is as follows:

First, focus the incident beam on the inside of the sample, and collect the OCT en face image sequence of the selected area inside the sample at equal intervals (the image number P of the image sequence is selected between 4-100, a natural number, and 20 is selected in this embodiment. ), a schematic diagram of OCT en face image sequence acquisition of a selected area inside the sample is shown in Figure 2, 201 represents the scanning objective, 202 represents the sample to be tested, 203 represents the sample stage, and 204 represents the selected area inside the sample. The length of the selected area inside the sample is equal to 0.8-1 times the original focal depth of the incident beam. In this embodiment, the original focal depth of the incident beam is 60 microns, and the length of the selected area inside the sample is taken as the original focal depth of the incident beam. 1 times, take 60 microns. As shown in Fig. 2(a), the incident beam is focused inside the sample, and a set of OCT en face image sequences are collected within the original focal depth range of the incident beam as a reference image sequence, and the reference image sequence belongs to OCT en face High-resolution image; then move the sample stage 203 along the specified direction, so that the position of the sample to be tested 202 is gradually moved away from or close to the scanning objective lens 201, and the focus position of the light beam inside the sample is changed, so that the selected area 204 inside the sample has different defocusing quantity. According to different defocus amount settings, M OCT en face low resolution image sequences under different defocus amounts are collected, and the OCT en face low resolution image sequences under different defocus amounts correspond to different resolutions. In this embodiment, we make the sample 202 gradually move away from the scanning objective lens 201, set M=4, and set the defocus amount to 60 μm, 120 μm, 180 μm and 240 μm, respectively, as shown in Figures 2(b)-2(e) respectively .

For the OCT en face low-resolution image sequences with different resolutions, the OCT en face image registration method is used in the reference image sequence to find and register the OCT en face high-resolution images at the corresponding positions, and four different defocusing amounts are established. datasets, each dataset contains 20 registered OCT en face low- and high-resolution image pairs.

The design of the OCT en face image registration method in the embodiment is shown in accompanying drawing 3, including;

First select the OCT en face low-resolution image and high-resolution image pair, the specific steps are as follows:

For each out-of-focus OCT en face low-resolution image extracted from the OCT en face low-resolution image sequence collected from a certain defocus position, select an OCT en face high-resolution image from the reference image sequence for pairing;

The specific method of obtaining an image pair: Preliminarily register a low-resolution OCT en face image with an out-of-focus image with each high-resolution OCT en face image in the reference image sequence using affine transformation; then, calculate the pre-registered image. The correlation coefficient r between the out-of-focus OCT en face low-resolution image and each OCT en face high-resolution image, from which the OCT en face high-resolution image with the highest correlation coefficient and the out-of-focus OCT en face low-resolution image are selected as paired images ;

The correlation coefficient r between the two images is calculated by the following formula:

(1)

(1)

代表高分辨图像的平均灰度值，

代表低分辨图像的平均灰度值。where f(x,y) and g(x,y) represent the gray value of the high-resolution image and low-resolution image respectively, x represents the abscissa of the image, y represents the ordinate of the image,

represents the average gray value of the high-resolution image,

Represents the average gray value of the low-resolution image.

Then, the pyramid registration algorithm is used to complete the fine multi-scale image registration on the initially registered OCT en face low-resolution image and high-resolution image pair. The specific implementation method is as follows:

The image is first divided into N×N image blocks (N=5 in the embodiment), and a two-dimensional normalized cross-correlation map of the corresponding image blocks is calculated.

The cross-correlation map (CCM) between the high-resolution image f and the low-resolution image g is defined as follows:

(2)

(2)

where u and v represent the abscissa and ordinate of the CCM, respectively.

The normalized cross-correlation map (nCCM) between the high-resolution image f and the low-resolution image g is calculated as follows:

(3)

(3)

是f的标准差，

是g的标准差，max表示求最大值，min表示求最小值，

是PPMCC的最大值，

是PPMCC的最小值。where PPMCC is the Pearson product-moment correlation coefficient (PPMCC), and cov() represents the covariance function,

is the standard deviation of f,

is the standard deviation of g, max indicates the maximum value, min indicates the minimum value,

is the maximum value of PPMCC,

is the minimum value of PPMCC.

The normalized cross-correlation plot is then fitted to a two-dimensional Gaussian function, defined as:

(4)

(4)

和

分别表示x和y方向上的标准差。二维高斯函数的峰值坐标就对应每个图像块的位移；5×5个图像块得到5×5个位移，把这些位移分别沿x和y方向插值到图像的像素个数，获得与原图像大小一致的位移图，利用该图像进行OCT en face低分辨图像和高分辨图像配准。如果位移图中的最大灰度值（或者容差）大于设定值，则重复附图3中的步骤 2.2.2-2.2.4。如果容差小于设定值，判断图像块大小是否小于设定的最小图像块大小，如果不小于设定值，则增加N值（赋值N=N+2）并执行附图3中的步骤2.2.1-2.2.5。如果小于设定的最小图像块大小，则完成图像配准，此时输入的低分辨图像和高分辨图像就达到了亚像素级配准。本实施例中容差的设定值为0.2；最小图像块大小为40×40（像素×像素）。where x ₀ and y ₀ represent the sub-pixel displacement values of the input image pair in two directions, respectively, A represents the gray value,

and

are the standard deviations in the x and y directions, respectively. The peak coordinate of the two-dimensional Gaussian function corresponds to the displacement of each image block; 5 × 5 image blocks get 5 × 5 displacements, and these displacements are interpolated to the number of pixels in the image along the x and y directions respectively, and the original image is obtained. Displacement map of the same size, which is used for OCT en face low-resolution image and high-resolution image registration. If the maximum gray value (or tolerance) in the displacement map is greater than the set value, repeat steps 2.2.2-2.2.4 in Figure 3. If the tolerance is less than the set value, judge whether the image block size is less than the set minimum image block size, if not less than the set value, increase the N value (assign N=N+2) and perform step 2.2 in Figure 3 .1-2.2.5. If it is smaller than the set minimum image block size, the image registration is completed, and the input low-resolution image and high-resolution image at this time achieve sub-pixel level registration. The set value of the tolerance in this embodiment is 0.2; the minimum image block size is 40×40 (pixel×pixel).

After image registration, 4 datasets of OCT en face low-resolution and high-resolution image pairs registered under different defocus amounts are established, each dataset contains 20 image pairs. Among them, the image size is 1000 × 1000 (pixel × pixel), which corresponds to a field of view of 3 mm × 3 mm. Divide the large image into image blocks with a size of 80 × 80 (pixel × pixel), remove the image blocks with less feature information when dividing, and then perform data expansion such as flipping, rotation, etc., to generate 3000 pairs of 80 × 80 (pixel × pixel). ) OCT en face low-resolution image and high-resolution image pair dataset. Finally, the training set, validation set and test set are distributed in a ratio of 3:1:1.

Step 2: Build the OCT en face image super-resolution model;

The specific implementation method is as follows:

In the construction of the OCT en face image super-resolution model, the deep learning method is used to learn the mapping relationship between the OCT en face high-resolution image and the OCT en face low-resolution image of different resolutions, and the digital refocusing of the OCT three-dimensional image is obtained to realize the OCT image. Depth of focus expansion.

In this embodiment, a generative adversarial neural network is used as the super-resolution model of deep learning (the image super-resolution model can also choose a convolutional neural network, a capsule network or a graph neural network). The generative adversarial neural network consists of a generator and a discriminator. The low-resolution image is input to the generator of the generative adversarial network to obtain the generated image. The generated image and the high-resolution image are input to the discriminator, and the discriminator is used to predict the real high-resolution image. Identify the probability that the image is more realistic than the generated image, and return the loss result to update the generator and discriminator weight parameters, and so on and so forth until the generative adversarial neural network converges.

The predicted probability of the discriminator is as follows:

(5)

(5)

为生成图像,

为高分辨图像，

表示判断高分辨图像真实的概率，

表示判断超分辨图像更真实的概率，C(x)为判别器的输出,

和

分别代表对所有生成数据和高分辨取平均值的操作。对抗损失结果由判别器生成，用于更新判别器的对抗损失函数

由公式(6)计算，更新生成器的对抗损失函数

由公式(7)计算；in

To generate images,

for high-resolution images,

represents the probability of judging the real high-resolution image,

Indicates the probability of judging that the super-resolution image is more realistic, C(x) is the output of the discriminator,

and

represent the operation of averaging over all generated data and high resolution, respectively. The adversarial loss result is generated by the discriminator and used to update the adversarial loss function of the discriminator

Calculated by Equation (6), the adversarial loss function of the generator is updated

Calculated by formula (7);

(6)

(6)

(7)

(7)

The calculation method of pixel loss function is shown in formula (8):

(8)

(8)

Where w and h represent the total number of pixels in the width and height directions of the image, respectively, and x and y represent the abscissa and ordinate of the image. Both w and h are equal to 80 in the examples.

The feature loss function measures the semantic difference between images through a pre-trained image classification network, using K. Simonyan et al. (Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition. in Proceedings of International Conference on Learning Representations. 2015, 1–14.) to extract receptive domain features from the pretrained VGG19 network described. VGG19 is a convolutional neural network consisting of 16 convolutional layers and 3 fully connected layers. The feature loss function is defined by formula (9):

(9)

(9)

表示在VGG19网络中第i层池化层后第j层卷积层输出的特征图，

和

是特征图的高和宽。在这里使用特征图

定义的

作为特征损失函数。in,

Represents the feature map output by the j-th convolutional layer after the i-th pooling layer in the VGG19 network,

and

are the height and width of the feature map. Use feature map here

Defined

as a feature loss function.

Adopt confrontation loss function to be the loss function of discriminator in the embodiment; The loss function of generator comprises confrontation loss function, pixel loss function and feature loss function, and its calculation method is such as formula (10):

(10)

(10)

The weighting parameters m, θ and η are used to control the trade-off between the three loss functions. In this embodiment, m=0.01, θ=1, and η=0.005.

The generator network structure in the example generative adversarial network is shown in FIG. 4, including a convolution module, a feature extraction module and a feature reconstruction module;

The convolution module includes a convolution layer Conv;

The feature extraction module includes 64 residual densely connected blocks RRDB Block. The structure of each residual dense connection block is shown in Figure 5. Each residual dense connection block consists of 23 dense connection blocks, and skip connections are added between each dense connection block; the dense connection block consists of five The convolutional layer Conv is composed of four LReLU layers, and in each block, the feature maps of each layer are concatenated with all previous features of the same scale.

The feature reconstruction module consists of two convolutional layers Conv and one LReLU layer to reconstruct the learned features into super-resolved images.

The size (k), number (n), and stride (s) of the convolution kernels corresponding to each convolutional layer in the generator network of the embodiment are marked in Figures 4 and 5.

The structure of the embodiment discriminator is shown in Fig. 6, including 8 convolutional layers Conv with 3 × 3 filter kernels. Except for the first layer, each convolutional layer is followed by a batch normalization layer BN and an LReLU layer. The convolutional layer is followed by two linear layers and an LReLU layer. The size (k), number (n), and stride (s) of convolution kernels corresponding to each convolutional layer are marked in Figure 6.

Using the OCT en face low-resolution image and high-resolution image datasets registered under 4 different defocus amounts, the generative adversarial neural network is trained respectively, and the OCT en face image super-resolution model under 4 different defocus amounts is obtained. The OCTen face image super-resolution models are all optimized by the Adam algorithm, and the hyperparameters are α = 0, β ₁ = 0.9, and β ₂ = 0.99. Set the number of training iterations to 150,000. The training and testing of the generative adversarial neural network is done using the deep learning framework Pytorch.

Step 3: realize digital refocusing of OCT images based on the constructed OCT en face image super-resolution model; the specific implementation method is as follows:

For the OCT 3D image data to be processed, first determine the focal position of the incident beam, divide the OCT 3D image to be processed according to the focal position, the original focal depth and the defocus amount, and obtain the OCT en face within the original focal depth range High-resolution image sequence and 4 groups (total 8 groups) of OCT enface low-resolution image sequences before and after the depth of focus range along the depth direction respectively; among them, the high-resolution OCT en face images do not need super-resolution processing; The defocus amounts corresponding to the four groups of OCT en face low-resolution image sequences in the depth direction are 30μm-90μm, 90μm-150μm, 150μm-210μm, and more than 210μm, respectively. Using the constructed OCT en face image super-resolution models at four corresponding defocus amounts (60μm, 120μm, 180μm and 240μm), respectively, four groups of OCT en face were divided into four groups of OCT en face before and after the focal depth range divided by different defocus amounts. The face low-resolution image sequence is subjected to super-resolution processing, and 4 sets of OCT en face super-resolution image sequences are obtained before and after the focal depth range. The lateral resolution of the OCT en face image is obtained, and finally, a set of OCT en face high-resolution image sequences within the focal depth range without super-resolution processing and the processed 8 sets of OCT en face super-resolution image sequences are respectively along the depth direction. The front of the focal depth range and the back of the focal depth range are re-stacked to form a refocusing OCT 3D image to achieve focal depth expansion.

As shown in FIG. 7 , FIG. 8 and FIG. 9 , the digital refocusing results of the OCT en face image of the present invention are presented. Fig. 7 is a low-resolution image of zebrafish OCT en face with a defocus amount of 240 μm, Fig. 8 is a high-resolution image of OCT en face obtained from the same sample in the range of focal depth, and Fig. 9 is the digital image of Fig. 7 output by the present invention Focused or super-resolved images, clear zebrafish structures can be seen from digitally refocused images. This shows that by the method of the present invention, the lateral resolution of the out-of-focus OCT en face image is improved, and the focal depth expansion of the OCT image is realized.

When the present invention is implemented, firstly, according to the defocus amount setting, the image registration algorithm is used to construct the OCT en face high-resolution image and the low-resolution image pair data set, then the OCT en face image super-resolution model is constructed, and the OCT en face is learned through the deep learning method. The mapping relationship between high-resolution images and OCT en face low-resolution images of different resolutions is based on the super-resolution model of OCT en face images at different defocusing amounts to obtain digital refocusing of OCT 3D images, and to expand the depth of focus of OCT images. The invention can realize the focal depth expansion of the optical coherence tomography image without increasing the hardware complexity of the OCT system, and has high speed and strong portability.

The above-mentioned embodiment is to illustrate the technical concept and characteristics of the present invention, and its purpose is to enable those who are familiar with the art to understand the content of the present invention and implement it, and cannot limit the protection scope of the present invention with this. Equivalent changes or modifications should be included within the protection scope of the present invention.

Claims (5)

1. a processing method for obtaining optical coherence tomography large depth of focus image, is characterized in that, comprises three steps: Step 1: Construct the OCT en face image dataset; The OCT en face image dataset refers to a dataset composed of pairs of OCT en face low-resolution images and high-resolution images registered under different defocus amounts; the specific construction method: Step 1.1: Focus the incident beam on the inside of the sample, and collect the OCT en face image sequence at equal intervals for the selected area inside the sample; the specific method is as follows: Move the sample stage along the specified direction, and collect the OCT en face image sequence of the selected area inside the sample when the incident beam is focused at M+1 positions, and the number of images collected in each image sequence is P; A set of OCT en face image sequences collected within the original focal depth range of the beam is taken as the reference image sequence, and the reference image sequence belongs to the OCT en face high-resolution images; the rest are collected from M positions that are not within the original focal depth range of the incident beam The M groups of image sequences are OCT en face low-resolution image sequences under different defocus amounts, and the OCT en face low-resolution image sequences under different defocus amounts correspond to different resolutions; The length of the selected area inside the sample is equal to 0.8-1 times the original focal depth of the incident beam; The M is a natural number between 2-8; The number P of images collected by each image sequence is a natural number between 4 and 100; Step 1.2: For the OCT en face low-resolution image sequences with different resolutions, use the OCT en face image registration method in the reference image sequence to find and register the OCT en face high-resolution images at the corresponding positions, and establish M different distances. Data sets under the focal volume, each data set contains P registered OCT en face low-resolution image and high-resolution image pairs; Step 2: Build the OCT en face image super-resolution model; Learning the mapping relationship between OCT en face high-resolution images and M groups of OCT en face low-resolution images with different resolutions through deep learning methods, and constructing M OCT en face image super-resolution models; Step 3: realize digital refocusing of OCT images based on the OCT en face image super-resolution model; Step 3.1: For the OCT 3D image data to be processed, first determine the focal position of the incident beam, divide the OCT 3D image to be processed according to the focal position, the original focal depth and the defocus amount, and obtain the OCT 3D image within the original focal depth range. The OCT enface high-resolution image sequence and each M group in the depth direction before and after the focal depth range, a total of 2M groups of OCT enface low-resolution image sequences; the OCT enface high-resolution image does not require super-resolution processing; Step 3.2: Use the M OCT en face image super-resolution models constructed in Step 2 to carry out M groups of OCT en face low-resolution image sequences before and after the focal depth range divided under different defocusing amounts, respectively. Image super-resolution processing, obtaining each M group of OCT en face super-resolution image sequences before and after the focal depth range, realizing image refocusing, and finally refocusing a group of OCT en face in the focal depth range without super-resolution processing. The resolved image sequences and the 2M groups of the OCT en face super-resolution image sequences obtained by processing are respectively re-stacked before and after the focal depth range in the depth direction to form a refocused OCT three-dimensional image to achieve focal depth expansion. 2. the processing method that obtains the optical coherence tomography large focal depth image according to claim 1, is characterized in that, described OCT en face image registration method comprises two steps: OCT en face low-resolution image and high-resolution image pair. Selection, fine registration of OCTen face low-resolution images and high-resolution images. 3. the processing method that obtains optical coherence tomography large focal depth image according to claim 2, is characterized in that, the selection method of described OCT en face low-resolution image and high-resolution image pair is, for from a certain defocus position For each defocused OCT en face low-resolution image extracted from the collected OCT en face low-resolution image sequence, select an OCT en face high-resolution image from the reference image sequence for pairing; The specific method for obtaining an image pair: preliminarily register a certain defocused OCT en face low-resolution image with each OCT en face high-resolution image in the reference image sequence; then, calculate the defocus after the preliminary registration The correlation coefficient r between the OCT enface low-resolution image and each of the OCT en face high-resolution images, from which the OCT en face high-resolution image with the highest correlation coefficient and the defocused OCT en face low-resolution image are selected as paired images; The correlation coefficient r between the two images is calculated by the following formula:

(1) where f(x,y) and g(x,y) represent the gray value of the high-resolution image and low-resolution image respectively, x represents the abscissa of the image, y represents the ordinate of the image,

represents the average gray value of the high-resolution image,

Represents the average gray value of the low-resolution image. 4. the processing method that obtains the optical coherence tomography large focal depth image according to claim 3, is characterized in that, the method for the preliminary registration of described OCT en face low-resolution image and high-resolution image comprises affine transformation method, rigid body transformation method, projection transformation method, nonlinear transformation method or moment and principal axis method. 5. the processing method that obtains optical coherence tomography large focal depth image according to claim 2, is characterized in that, the fine registration method of described OCT en face low-resolution image and high-resolution image comprises pyramid registration method, wavelet transform Registration method, maximum mutual information registration method and map registration method.

Images