Method for analyzing oocyte maturation

Abstract

The present application relates to the technical field of computer vision, and particularly relates to an oocyte maturity analysis method, device and computer storage medium, the analysis method comprises the following steps: S1, classifying oocyte images into oocyte images with granulosa cells removed and oocyte images without granulosa cells removed; further classifying into mature and immature; S2, data augmentation is performed on the immature; S3, the oocyte images are segmented into non-overlapping image blocks, through four-stage processing, the feature maps of each image block in multiple dimensions are obtained; the weight scores of the feature maps in different dimensions are calculated in the dimension direction; the probability value of the category to which the image to be predicted belongs is output through a full connection layer, and the analysis of oocyte maturity is completed; the method of the present application has higher recognition preparation rate, faster speed, stronger device mobility, and has a significant effect on recognizing oocyte images before granulosa cells are removed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer vision, in particular to an oocyte maturity analysis method. BACKGROUND

[0002] The number of infertile patients in China reaches 40 million, and it has a trend of increasing year by year. Obtaining oocytes is the most important prerequisite for assisted reproductive technology. The quality of oocytes reflects the potential of internal development, and plays a crucial role in fertilization and subsequent embryonic development. In the process of assisted reproductive technology, oocyte maturity analysis needs to be performed multiple times.

[0003] Oocytes are very valuable and rare for women, so the oocyte acquisition in assisted reproductive technology is very cautious. Only mature oocytes can be normally fertilized when artificial reproduction is performed. However, not all oocytes taken out are mature. Mature oocytes have a spherical structure surrounded by uniform zonapellucida, uniform cytoplasm, no inclusions, and appropriate polar body size, which is called MII phase. The immature oocytes, i.e. the cell membrane maturation is not synchronized with the cytoplasm maturation, will break the cell membrane when single sperm injection is performed, resulting in oocyte death. Immature oocytes are shown in Figure 1 and Figure 2 and Figure 3

[0004] At present, the determination of oocyte maturity is performed by manually judging whether there are cumulus granulosa cells in the image formed under an inverted microscope and performing rough morphological evaluation. The diagnosis obtained for the same image is different due to the limitation of personal experience and subjective influence. The morphology cannot be completely objectively and effectively evaluated, so an intelligent software is urgently needed to effectively and quickly realize the quantitative and automatic analysis of oocyte maturity, complete the heavy image interpretation work, and ensure that the evaluation results are consistent, objective and repeatable. The doctor can spend more effort on the development of treatment plan and research, and realize the rational allocation of resources. SUMMARY

[0005] The present application provides an oocyte maturity analysis method capable of better recognizing the image of oocytes without removing granulosa cells, which comprises the following steps:

[0006] S1, pre-processing and classifying the oocyte image into oocyte images with removed granulosa cells and oocyte images without removed granulosa cells;

[0007] ​According to the oocyte morphology rating information, the oocyte images with removed granulosa cells are divided into mature oocyte images with removed granulosa cells and immature oocyte images with removed granulosa cells, and the oocyte images without removed granulosa cells are divided into mature oocyte images without removed granulosa cells and immature oocyte images without removed granulosa cells;

[0008] The mature oocyte images with removed granulosa cells and the mature oocyte images without removed granulosa cells jointly form mature oocyte images, and the immature oocyte images with removed granulosa cells and the immature oocyte images without removed granulosa cells jointly form immature oocyte images;

[0009] S2, data augmentation is performed on the immature oocyte images with removed granulosa cells and the immature oocyte images without removed granulosa cells;

[0010] The immature oocyte images with removed granulosa cells after data augmentation and the mature oocyte images with removed granulosa cells jointly form an oocyte image data set with removed granulosa cells;

[0011] The immature oocyte images without removed granulosa cells after data augmentation and the mature oocyte images without removed granulosa cells jointly form an oocyte image data set without removed granulosa cells;

[0012] S3, a first image in the oocyte image data set with removed granulosa cells is divided into non-overlapping first image blocks, and each first image block is characterized by a concatenation of all pixel RGB values; four stages are constructed, and through the processing of the four stages, a multi-dimensional feature map of each first image block is obtained;

[0013] Based on the channel attention module of the convolutional network architecture, the weight scores of the feature maps in different dimensions are calculated in the dimension direction; the probability value of the category to which the first to-be-predicted image belongs is output through the fully connected layer, and the analysis of the oocyte maturity corresponding to the first to-be-predicted image is completed;

[0014] A second image in the oocyte image data set without removed granulosa cells is divided into non-overlapping second image blocks, and each second image block is characterized by a concatenation of all pixel RGB values; four stages are constructed, and through the processing of the four stages, a multi-dimensional feature map of each second image block is obtained;

[0015] Based on the channel attention module of the convolutional network architecture, the weight scores of the feature maps in different dimensions are calculated in the dimension direction; the probability value of the category to which the second to-be-predicted image belongs is output through the fully connected layer, and the analysis of the oocyte maturity corresponding to the second to-be-predicted image is completed.

[0016] Preferably, the data augmentation manner comprises at least one of random rotation, contrast stretch, histogram stretch, sharpening, and brightness change.

[0017] Preferably, the difference between the data amount of the immature degranulated oocyte image after data augmentation and the data amount of the mature degranulated oocyte image is not more than 1% to 5%; the difference between the data amount of the immature degranulated oocyte image after data augmentation and the data amount of the mature degranulated oocyte image is not more than 1% to 5%.

[0018] Preferably, the step S3 comprises that the size of the first image or the second image is HxWx3, H and W represent the height and width of the first image or the second image respectively, the size of each first image block or second image block is 4x4, and after division, the height and width of the first image or the second image become 4Hx4W respectively. The depth of the first image or the second image becomes 4x4x3=48.

[0019] Preferably, the four stages comprise a first stage, a second stage, a third stage, and a fourth stage.

[0020] The processing of the first stage comprises linearly transforming the channel data of each pixel of the first image or the second image through a linear embedding layer, without changing the height and width of the first image or the second image, and changing the dimension (C) of the input vector into a preset value; after the linear embedding layer, the size of the first image or the second image becomes The first image or the second image is input into a moving window self-attention transformation module to realize global modeling; before the processing of the second stage, the processing of the third stage, and the processing of the fourth stage, the first image or the second image is down-sampled through the block fusion layer respectively.

[0021] Preferably, the moving window self-attention transformation module is obtained by connecting a window multi-head self-attention mechanism module and a moving window multi-head self-attention mechanism module in series.

[0022] Preferably, the input of the first image or the second image into the moving window self-attention transformation module realizes global modeling, which includes: first, using the moving window self-attention transformation module, the input first image or the second image is divided into a plurality of non-overlapping windows, the self-attention score is calculated in the window, and the obtained self-attention score is combined using multi-head to make the model understand the input features from different angles and generate a multi-head self-attention feature map; then the first image or the second image is sent into the moving window multi-head self-attention mechanism module, and the original window and the moved window are partially overlapped through the moving window operation, the communication between the windows is realized, and the feature multi-head self-attention score between different regions is calculated to realize global modeling.

[0023] After the first stage processing, the feature map size of the first image or the second image is After the second stage processing, the feature map size of the first image or the second image is After the third stage processing, the feature map size of the first image or the second image is After the fourth stage processing, the feature map size of the first image or the second image is

[0024] The application also provides an oocyte maturity analysis device, comprising a processor, which is used to realize the steps of the oocyte maturity analysis method described above when executing the computer program stored in the memory.

[0025] The application also provides a computer readable storage medium, which stores an oocyte maturity analysis method program, and the steps of the oocyte maturity analysis method described above are realized when the processor executes the oocyte maturity analysis method program.

[0026] The oocyte maturity analysis method provided by the application has higher judgment accuracy, and the judgment accuracy of the oocyte maturity of the oocyte without removing granulosa cells is more than 80%, and the judgment accuracy of the oocyte maturity of the oocyte with removed granulosa cells is more than 90%; the fusion network architecture is used, and compared with the traditional convolution network architecture, the recognition accuracy of the oocyte maturity is improved by about 20%; and the recognition accuracy of the oocyte image without removing granulosa cells is also improved by about 20%; the computer vision technology is used for judgment, the time of the doctors and other professionals is saved, and the doctors are helped to quickly perform medical diagnosis; the judgment speed is fast, and the time required for judging the oocyte maturity of one oocyte is not more than one minute; and the device has strong mobility, and all tasks can be completed by one notebook computer without the need of a particularly complex device.

[0027] In addition, the oocyte maturity analysis method provided by the application can identify immature oocytes from the oocyte image before the granulosa cells are removed, and can complete subsequent in-vitro culture to the mature stage, so as to facilitate subsequent assisted reproduction, and greatly reduce the psychological and physical harm to patients; effectively solve the problem that if the cell is identified as an immature oocyte, the oocyte cannot be cultured subsequently due to the removal of granulosa cells, and new oocytes need to be extracted from the female patient, which increases the pain of the patient. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 is an electron microscope image of a generally mature oocyte.

[0029] Figure 2 and Figure 3 is an electron microscope image of a generally immature oocyte.

[0030] Figure 4 is an augmented image of the oocyte image data of the immature oocyte without removing the granulosa cells in the specific embodiment of the application.

[0031] Figure 5 is an augmented image of the oocyte image data of the immature oocyte without removing the granulosa cells in the specific embodiment of the application.

[0032] Figure 6 is a flowchart of image recognition in the specific embodiment of the application.

[0033] Figure 7 is a before-and-after effect diagram of the first oocyte maturity recognition without removing the granulosa cells in the specific embodiment of the application.

[0034] Figure 8 is a before-and-after effect diagram of the second oocyte maturity recognition without removing the granulosa cells in the specific embodiment of the application.

[0035] Figure 9 is a before-and-after effect diagram of the first oocyte maturity recognition without removing the granulosa cells in the specific embodiment of the application.

[0036] Figure 10 is a before-and-after effect diagram of the second oocyte maturity recognition without removing the granulosa cells in the specific embodiment of the application. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solutions and advantages of the application clearer, the application will be further described in detail below in combination with the drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the application, and do not constitute a limitation on the application.

[0038] The method for analyzing oocyte maturity provided in the embodiments of the present application comprises the following steps:

[0039] S1, pre-process and classify the oocyte images into oocyte images with granulosa cells removed and oocyte images without granulosa cells removed; according to the oocyte morphology rating information, divide the oocyte images with granulosa cells removed into mature oocyte images with granulosa cells removed and immature oocyte images with granulosa cells removed, and divide the oocyte images without granulosa cells removed into mature oocyte images without granulosa cells removed and immature oocyte images without granulosa cells removed; the mature oocyte images with granulosa cells removed and the mature oocyte images without granulosa cells removed together form mature oocyte images, and the immature oocyte images with granulosa cells removed and the immature oocyte images without granulosa cells removed together form immature oocyte images.

[0040] In the specific embodiments, a program can be written according to the provided oocyte images and text information, wherein the text information includes image number, pre-ovum removal image information, and oocyte morphology rating information. First, it is determined whether the oocyte images without granulosa cells removed are contained in the same image number according to the pre-ovum removal image information. After this operation, a folder corresponding to the image number is generated, which contains the oocyte images of the corresponding image number and is divided into two categories, one of which includes images with granulosa cells removed, and the other of which includes images without granulosa cells removed. Then, according to the oocyte morphology rating information, the two categories of cell images are further subdivided, the oocyte images with granulosa cells removed are divided into mature oocyte images with granulosa cells removed and immature oocyte images with granulosa cells removed, and the oocyte images without granulosa cells removed are divided into mature oocyte images without granulosa cells removed and immature oocyte images without granulosa cells removed. Finally, the oocyte images are divided into mature oocytes (MII phase) and immature oocytes (non-MII phase).

[0041] S2, data augmentation is performed on the immature oocyte images with granulosa cells removed and the immature oocyte images without granulosa cells removed; the data-augmented immature oocyte images with granulosa cells removed and the mature oocyte images with granulosa cells removed together form an oocyte image data set with granulosa cells removed; and the data-augmented immature oocyte images without granulosa cells removed and the mature oocyte images without granulosa cells removed together form an oocyte image data set without granulosa cells removed.

[0042] In a specific embodiment, the data augmentation method includes at least one of random rotation, contrast stretching, histogram stretching, sharpening, and brightness variation. In a preferred embodiment, the data augmentation method includes random rotation, contrast stretching, histogram stretching, sharpening, and brightness variation. By using all kinds of data augmentation methods, the immature cell images of immature oocytes obtained from the removed and non-removed granulosa cells are transformed, so that the original immature oocyte image data (including immature oocyte images of removed granulosa cells and immature oocyte images of non-removed granulosa cells) is augmented as much as possible. For example, Figure 4 and Figure 5 As shown in the figures, they are the augmented images of immature oocyte images of non-removed granulosa cells and immature oocyte images of removed granulosa cells, respectively.

[0043] In a specific embodiment, the immature oocyte (non-MⅡ phase) data in the actually obtained oocyte data is relatively scarce, which causes the positive and negative samples to be seriously unbalanced, which is not conducive to the subsequent model training. Therefore, the immature oocyte images are augmented to make the data amount of the augmented immature oocyte images the same as or close to that of the mature oocyte images. Specifically, the difference between the data amount of the augmented immature oocyte images of removed granulosa cells and the data amount of the mature oocyte images of removed granulosa cells is not more than 1% to 5%. The difference between the data amount of the augmented immature oocyte images of non-removed granulosa cells and the data amount of the mature oocyte images of non-removed granulosa cells is not more than 1% to 5%, so as to meet the subsequent training requirements.

[0044] S3, in order to distinguish the images in the oocyte image dataset with granulosa cells removed from the images in the oocyte image dataset without granulosa cells removed, the images in the oocyte image dataset with granulosa cells removed are defined as first images, and the images in the oocyte image dataset without granulosa cells removed are defined as second images; in this step, the first images in the oocyte image dataset with granulosa cells removed are divided into non-overlapping first image blocks, and each first image block is characterized by a concatenation of all pixel RGB values; four stages are constructed, and through the processing of the four stages, the multi-dimensional feature maps of each first image block are obtained; based on the channel attention module of the convolutional network architecture, the weight scores of the feature maps in different dimensions are calculated in the dimension direction; the probability value of the category to which the first predicted image belongs is output through the fully connected layer, and the analysis of the oocyte maturity corresponding to the first predicted image is completed; the second images in the oocyte image dataset without granulosa cells removed are divided into non-overlapping second image blocks, and each second image block is characterized by a concatenation of all pixel RGB values; four stages are constructed, and through the processing of the four stages, the multi-dimensional feature maps of each second image block are obtained; based on the channel attention module of the convolutional network architecture, the weight scores of the feature maps in different dimensions are calculated in the dimension direction; the probability value of the category to which the second predicted image belongs is output through the fully connected layer, and the analysis of the oocyte maturity corresponding to the second predicted image is completed.

[0045] In the specific implementation, the size of the first image or the second image is HxWx3, H and W represent the height and width of the first image or the second image respectively, the size of each first image block or second image block is 4x4, and after division, the height and width of the first image or the second image become The depth of the first image or the second image becomes 4x4x3=48; the four stages include a first stage, a second stage, a third stage, and a fourth stage; the processing of the first stage includes linearly transforming the channel data of each pixel of the first image or the second image through a linear embedding layer, without changing the height and width of the first image or the second image, and changing the dimension (C) of the input vector to a preset value; after the linear embedding layer, the size of the first image or the second image becomes The first image or the second image is input into a moving window self-attention transformation module to realize global modeling; the moving window self-attention transformation module is obtained by connecting a window multi-head self-attention mechanism module and a moving window multi-head self-attention mechanism module in series; first, the moving window self-attention transformation module is used to divide the input first image or second image into a plurality of non-overlapping windows, and self-attention scores are calculated in the windows; the obtained self-attention scores are combined using multi-head to enable the model to understand the input features from different angles and generate a multi-head self-attention feature map; then, the first image or the second image is sent to the moving window multi-head self-attention mechanism module, and a moving window operation is performed to make the original window partially overlap with the moved window, realize the communication between the windows, and thus calculate the multi-head self-attention scores of the features between different regions to realize global modeling; before being processed by the second stage, the third stage and the fourth stage, the first image or the second image is respectively down-sampled by the block fusion layer. After the first stage processing, the feature map size of the first image or the second image is After the second stage processing, the feature map size of the first image or the second image is After the third stage processing, the feature map size of the first image or the second image is After the fourth stage processing, the feature map size of the first image or the second image is

[0046] In the specific implementation, the oocyte maturity analysis method provided by the application recognizes the maturity of oocytes, and the specific image recognition process is as shown in Figure 6 The oocyte image without granulosa cells and the oocyte image with granulosa cells are respectively input into the network for training by adopting the fusion architecture of the moving window-based self-attention mechanism network architecture and the convolution-based network architecture; taking the training process of one of the images as an example, first, the image obtained after data augmentation is divided into non-overlapping patches; the feature of each patch is the concatenation of all pixel RGB values, and four stages are constructed, and each stage is a similar repeating unit; first, the image partition layer is used to divide the image with an input size of HxWx3 into a set of non-overlapping patches, H and W represent the height and width of the image, and each patch has a size of 4x4; after the division, the height and width of the image become The depth becomes 4x4x3=48, and then the channel data of each pixel is linearly transformed through a linear embedding layer. This process does not change the height and width of the image, but changes the dimension (C) of the input vector to a preset value. After passing through this layer, the shape of the image becomes The image is input into the first mobile window self-attention transformation module (SwinTransformerBlock), which is composed of two self-attention modules, a window multi-head self-attention mechanism module (W-MSA) and a shifted window multi-head self-attention mechanism module (SW-MSA). First, W-MSA is used to divide the input image into multiple non-overlapping windows, calculate the self-attention scores in the given window, and reduce the computational complexity. Then, multi-head is used to combine the obtained attention scores, so that the model can understand the input features from different angles, and finally generate a multi-head self-attention feature map. Then, the image is input into SW-MSA, which is operated by a shifted window, so that the original window and the shifted window have overlapping parts, realizing the communication between the windows, calculating the feature multi-head attention scores between different regions, and achieving the effect of global modeling. At this time, the stage1 stage processing is completed, and the output feature map size is still Before stage2-stage4 operation, an image block merging layer is used for downsampling, so that the height and width of the feature map from the previous stage are halved, and the dimension is doubled. Therefore, after stage2-stage4 processing, the feature map size becomes Finally, considering that the obtained feature map is multi-dimensional, it is further input into a channel attention module based on a convolutional network architecture to calculate the weight scores of the feature map in different dimensions in the dimension direction, so as to better capture the importance of the features in different dimensions. Finally, a fully connected layer is used to output the probability value of the predicted image belonging to the category, that is, the probability value of the predicted image belonging to the mature granulosa cell image or the probability value of the predicted image belonging to the immature granulosa cell image, and the analysis of the maturity of the oocyte is completed.

[0047] The application also provides an oocyte maturity analysis device, which comprises a processor for executing the computer program stored in the memory to realize the steps of the oocyte maturity analysis method described above.

[0048] The application also provides a computer readable storage medium, wherein the computer readable storage medium stores an oocyte maturity analysis method program, and the oocyte maturity analysis method program, when executed by a processor, implements the steps of the oocyte maturity analysis method.

[0049] The following proves the actual recognition effect of the oocyte maturity analysis method provided by the application on oocyte maturity through a specific experimental process.

[0050] The original oocyte images without removing granulosa cells include 569 mature oocyte images and 77 immature oocyte images, 14 of which are selected as test set, including 5 mature oocyte images and 9 immature oocyte images, and the rest of the image data is used as training data.

[0051] The original oocyte images without removing granulosa cells include 569 mature oocyte images and 77 immature oocyte images, 14 of which are selected as test set, including 5 mature oocyte images and 9 immature oocyte images, and the rest of the image data is used as training data.

[0052] The data-augmented oocyte images without removing granulosa cells include 2067 mature oocyte images and 2079 immature oocyte images.

[0053] The data-augmented oocyte images without removing granulosa cells include 3728 mature oocyte images and 3801 immature oocyte images.

[0054] The oocyte images in the test data set of oocytes without removing granulosa cells and oocytes removing granulosa cells are identified by using the commonly used vgg19, resnet152, inception_v3 and mobilenet_v3 analysis methods and the oocyte maturity analysis method provided by the application, and the identification accuracy results are shown in Tables 1 and 2.

[0055] Table 1 Identification accuracy of oocyte without removing granulosa cells

[0056]

[0057] The oocyte images in the test data set of oocytes without removing granulosa cells and oocytes removing granulosa cells are identified by using the commonly used vgg19, resnet152, inception_v3 and mobilenet_v3 analysis methods and the oocyte maturity analysis method provided by the application, and the identification accuracy results are shown in Tables 1 and 2. Figure 7 and Figure 8respectively, are the effect pictures before and after the first and second granulosa cell unremoved oocyte maturity recognition in the embodiment of the present application, from the pictures, it can be seen that the analysis method of the present application can well recognize the oocyte maturity of the granulosa cell unremoved image.

[0058] Table 2 granulosa cell oocyte recognition accuracy

[0059]

[0060] The oocyte image test set data of the granulosa cell unremoved oocyte image has a total of 14 images, from the data in Table 2, compared with the traditional image recognition method based on the convolutional neural network architecture, the recognition accuracy of the oocyte maturity is improved by about 20% by using the analysis method of the present application; in addition, Figure 9 and Figure 10 respectively, are the effect pictures before and after the first and second granulosa cell unremoved oocyte maturity recognition in the embodiment of the present application, from the pictures, it can be seen that the analysis method of the present application can well recognize the oocyte maturity of the granulosa cell unremoved image.

[0061] The oocyte maturity analysis method provided by the present application has higher judgment accuracy, the oocyte maturity judgment accuracy of the granulosa cell unremoved oocyte is more than 80%, and the oocyte maturity judgment accuracy of the granulosa cell removed oocyte is more than 90%; the fusion network architecture is used, compared with the traditional convolutional network architecture, the recognition accuracy of the oocyte maturity is improved by about 20%; and the recognition accuracy of the oocyte image of the granulosa cell unremoved oocyte can also be improved by about 20%; the computer vision technology is used for identification, the time of the doctors and other professional personnel is saved, and the doctors are helped to quickly perform medical diagnosis; the judgment speed is fast, and the time required for judging the oocyte maturity of one oocyte is not more than one minute; and the equipment has strong mobility, and all tasks can be completed by one notebook computer, without the need of special complex equipment.

[0062] In addition, by using the oocyte maturity analysis method provided by the present application, the non-mature oocyte can be recognized from the oocyte image before the granulosa cell is removed, and the subsequent in vitro culture to the mature period can be completed, so as to assist the subsequent assisted reproduction, greatly reducing the psychological and physical harm to the patient; effectively solve the problem that once the cell is recognized as a non-mature oocyte, the oocyte cannot be cultured subsequently due to the removal of the granulosa cell, and a new oocyte needs to be extracted from the female patient, increasing the pain of the patient.

[0063] Although the embodiments of the present application have been shown and described above, it is understood that the above-described embodiments are exemplary and are not to be construed as limiting the present application. Those skilled in the art can make various changes, modifications, replacements and variations of the above-described embodiments within the scope of the present application.

[0064] The above detailed description of the application is not to be construed as limiting the scope of the application. Any other corresponding changes and modifications made according to the technical concept of the present application should be included in the scope of the protection of the present application.

Claims

1. A method for analyzing oocyte maturity, characterized in that: The analysis method comprises the steps of: S1, pre-processing and classifying the oocyte image, into an oocyte image with granulosa cells removed and an oocyte image without granulosa cells removed; According to the oocyte morphology rating information, the oocyte image with granulosa cells removed is divided into a mature oocyte image with granulosa cells removed and an immature oocyte image with granulosa cells removed, and the oocyte image without granulosa cells removed is divided into a mature oocyte image without granulosa cells removed and an immature oocyte image without granulosa cells removed; S2, data augmentation is performed on the immature oocyte image with granulosa cells removed and the immature oocyte image without granulosa cells removed; The immature oocyte image with granulosa cells removed after data augmentation and the mature oocyte image with granulosa cells removed form an oocyte image data set with granulosa cells removed; The immature oocyte image without granulosa cells removed after data augmentation and the mature oocyte image without granulosa cells removed form an oocyte image data set without granulosa cells removed; S3, the first image in the oocyte image data set with granulosa cells removed is divided into non-overlapping first image blocks, and the feature of each first image block is the concatenation of all pixel RGB values; four stages are constructed, including a first stage, a second stage, a third stage and a fourth stage; Through the processing of the four stages, the multi-dimensional feature map of each first image block is obtained; Based on the channel attention module of the convolutional network architecture, the weight score of the feature map in different dimensions is calculated in the dimension direction; the probability value of the category to which the first to-be-predicted image belongs is output through the fully connected layer, and the analysis of the oocyte maturity corresponding to the first to-be-predicted image is completed; The second image in the oocyte image data set without granulosa cells removed is divided into non-overlapping second image blocks, and the feature of each second image block is the concatenation of all pixel RGB values; Four stages are constructed, and through the processing of the four stages, the multi-dimensional feature map of each second image block is obtained; After the first stage processing, the feature map size of the first image or the second image is After the second stage processing, the feature map size of the first image or the second image is After the third stage processing, the feature map size of the first image or the second image is After the fourth stage processing, the feature map size of the first image or the second image is ; Based on the channel attention module of the convolutional network architecture, the weight score of the feature map in different dimensions is calculated in the dimension direction; the probability value of the category to which the second to-be-predicted image belongs is output through the fully connected layer, and the analysis of the oocyte maturity corresponding to the second to-be-predicted image is completed.

2. The oocyte maturation analysis method according to claim 1, wherein The data augmentation method includes at least one of random selection, rotation, contrast stretching, histogram stretching, sharpening, and brightness change.

3. The method of claim 1, wherein the step of determining the maturity of the oocyte comprises the step of: The difference between the data amount of the immature oocyte image with granulosa cells removed after data augmentation and the mature oocyte image with granulosa cells removed is not more than 1%~5%; the difference between the data amount of the immature oocyte image without granulosa cells removed after data augmentation and the mature oocyte image without granulosa cells removed is not more than 1%~5%. ​ 4. The method of claim 1, wherein the step of determining the maturity of the oocyte comprises the step of: In the step S3, the size of the first image or the second image is H × W × 3, H and W respectively represent the height and the width of the first image or the second image, the size of each of the first image block or the second image block is 4x4, after the division, the height and the width of the first image or the second image respectively become , and the depth of the first image or the second image becomes 4x4x3=48. ​ 5. The method of claim 4, wherein the step of determining the maturity of the oocyte comprises the step of: The first stage of processing includes linearly transforming channel data of each pixel of the first image or the second image through a linear embedding layer, without changing height and width of the first image or the second image, changing dimension (C) of an input vector into a preset value; after the linear embedding layer, size of the first image or the second image becomes ; the first image or the second image is input into a moving window self-attention transformation module to realize global modeling; before the second stage of processing, the third stage of processing and the fourth stage of processing, the first image or the second image respectively undergoes down-sampling through the block fusion layer. ​ 6. The oocyte maturity analysis method of claim 5, wherein, The moving window self-attention transformation module is obtained by concatenating a window multi-head self-attention mechanism module and a moving window multi-head self-attention mechanism module.

7. The method of claim 6, wherein the step of determining the maturity of the oocyte comprises the step of: The inputting the first image or the second image into the moving window self-attention transformation module to realize global modeling comprises: first, using the moving window self-attention transformation module to divide the input first image or second image into a plurality of non-overlapping windows, calculating self-attention scores in the windows, and using multi-head to combine the obtained self-attention scores to enable the model to understand the input features from different angles and generate multi-head self-attention feature maps; then, the first image or the second image is sent into the moving window multi-head self-attention mechanism module, a moving window operation is performed to make the original window partially overlap with the moved window, realize the mutual communication between the windows, calculate the multi-head self-attention scores of the features between different regions, and realize global modeling. ​ 8. An oocyte maturation degree analysis apparatus characterized by comprising: The computer program product comprises a computer readable storage medium storing a computer program, and the computer program is executed by a processor to implement the steps of the oocyte maturity analysis method according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores an oocyte maturity analysis method program, and the oocyte maturity analysis method program is executed by a processor to implement the steps of the oocyte maturity analysis method according to any one of claims 1 to 7.

Images