Method and device for evaluating stripping quality of cumulus cells

Abstract

The invention relates to the technical field of biomedical treatment, and discloses a cumulus cell stripping quality assessment method and device.The method comprises the steps that on different focusing layers, two focusing images of an ovum to be assessed after cumulus cells are stripped are collected, and the positions of all cumulus cells and the total number of the cumulus cells are recognized; summing the total number of cumulus cells in the two focused images to obtain an initial number; intercepting each cumulus cell in the two focused images, and pairing each cumulus cell in the first focused image and each cumulus cell in the second focused image to obtain a plurality of cumulus cell pairs; identifying the logarithm of the cumulus cell pairs of which the two cumulus cells are the same cumulus cell, and based on the position of each cumulus cell, rejecting the cumulus cell pairs of which the position difference is greater than a preset distance to obtain a possible cell logarithm; and calculating the difference between the initial number and the possible cell logarithm to obtain the number of cumulus cells around the to-be-evaluated ovum, and obtaining the cumulus cell stripping quality of the to-be-evaluated ovum.

Description

Technical Field

[0001] This invention relates to the field of biomedical technology, and in particular to a method and apparatus for assessing the quality of cumulus cell removal. Background Technology

[0002] Statistics show that due to factors such as delayed first marriage and childbearing age and irregular lifestyles, infertility now affects 8-12% of the global population of reproductive age. Since the birth of the world's first test-tube baby, Assisted Reproductive Technology (ART) has gradually become the most effective method for treating infertility. Cumulus cell removal quality assessment refers to the clinician repeatedly blowing and aspirating the oocyte under a stereomicroscope using a microtube, while simultaneously assessing the removal of the cumulus cells surrounding the oocyte. If the cumulus cells are not completely removed, the remaining cells may enter the subsequent ART procedure with the oocyte, causing adverse effects and even reducing the success rate of the ART procedure. Therefore, cumulus cell removal quality assessment is an essential part of ART.

[0003] Existing techniques mainly include manual observation and image thresholding. Manual observation relies on clinicians observing under a stereomicroscope with limited depth of field and making subjective assessments. Due to the limited depth of field, doctors often cannot fully observe the removal of all cumulus cells on the surface of the egg or embryo. If there are many eggs, subjective differences among doctors can lead to significant errors in assessing the removal quality. Furthermore, some researchers have attempted to use image thresholding to segment and identify the cumulus cell outline and calculate the area of ​​the outline region to assess removal quality. However, because the contrast between the egg and cumulus cells in the microscopic image is similar, image thresholding is less effective at segmenting and identifying the cumulus cell outline, thus failing to accurately assess the quality of cumulus cell removal.

[0004] In summary, manual observation, relying on limited visual information under a microscope to directly observe the removal of cumulus cells, is highly subjective and often depends on the physician's extensive experience and knowledge, leading to significant assessment errors. Furthermore, image thresholding is sensitive to noise and brightness variations in images, potentially causing instability in the segmentation and recognition of cumulus cell contours, thus reducing the accuracy of the removal quality assessment. Additionally, image thresholding may fail to effectively segment and recognize some complex cumulus cell contours. Moreover, existing methods for assessing cumulus cell removal quality cannot quantify the removal quality. Summary of the Invention

[0005] Therefore, the technical problem to be solved by the present invention is to overcome the problems of large subjective evaluation error or unstable segmentation and recognition in the prior art due to reliance on manual recognition or image threshold resolution.

[0006] To address the aforementioned technical problems, this invention provides a method for assessing the quality of cumulus cell removal, comprising: Oocytes obtained after cumulus cell removal were used as oocytes for evaluation; Two focused images of the oocyte to be evaluated were acquired at different focusing levels, and the location of all cumulus cells and the total number of cumulus cells in each focused image were identified. The initial number is obtained by summing the total number of cumulus cells in the first focused image and the total number of cumulus cells in the second focused image; Each cumulus cell in two focused images is extracted, and each cumulus cell in the first focused image is paired with each cumulus cell in the second focused image to obtain multiple cumulus cell pairs; Obtain the number of cumulus cell pairs in which the two cumulus cells are the same cumulus cell, and based on the position of each cumulus cell, remove cumulus cell pairs whose positions differ by a preset distance to obtain the possible number of cell pairs; The difference between the initial number and the possible number of cell pairs is calculated to obtain the number of cumulus cells surrounding the oocyte to be evaluated; The quality of cumulus cell removal of the oocyte is obtained based on the number of cumulus cells surrounding the oocyte to be evaluated and the preset quality grade.

[0007] Preferably, two focused images of the oocyte to be evaluated are acquired after the cumulus cells have been removed, including: The first-focus image of the oocyte to be evaluated was acquired using a microscope at the first focal depth. Adjust the depth of focus and acquire a second focused image of the oocyte to be evaluated at the second depth of focus.

[0008] Preferably, a deep learning detection model is constructed based on the YOLO model, and the deep learning detection model is used to identify the location of all cumulus cells and the total number of cumulus cells in each focused image, including: By introducing the Res2Net-COT residual structure into the backbone network of the YOLO model, multi-scale feature extraction is performed on the input focused image to obtain multi-scale cell morphology feature maps. A bidirectional feature pyramid structure is introduced into the neck network of the YOLO model to perform top-down and bottom-up feature fusion on the input multi-scale cell morphology feature map to obtain the fused feature map. The fused feature map is input into the detector head of the YOLO model for prediction, resulting in the bounding box for each cumulus cell; Based on the bounding boxes of all cumulus cells, the total number of cumulus cells and the location of each cumulus cell are obtained.

[0009] Preferably, a deep learning detection model is used to obtain the location of all cumulus cells and the total number of cumulus cells in the focused image.

[0010] Preferably, obtaining the training set for the deep learning detection model includes: Multiple oocytes were obtained, and two focused images were acquired for each oocyte; The number of focused images is increased by using image enhancement methods including rotation, horizontal flipping, vertical flipping, and contrast adjustment. All focused images are resized to a preset pixel size and manually labeled to obtain bounding boxes indicating the location of each cumulus cell in each focused image. The focused images and their corresponding bounding boxes are used as the training set.

[0011] Preferably, a deep learning classification model is constructed based on Siamese or SwAV, and the deep learning classification model is used to obtain the number of cumulus cell pairs in which two cumulus cells are the same cumulus cell, including: A spatial pyramid pooling layer structure is introduced into the feature extraction network of the model. The bounding box images of cumulus cells in two focused images are paired as input to extract the features of cumulus cells in each bounding box image and convert them into feature vectors of fixed dimensions. A feature band structure is introduced into the feature contrast network of the model to map the feature vectors to a high-dimensional feature space, obtain the corresponding denoised feature map, and calculate the contrast loss function value between each pair of bounding box images as the feature similarity of a pair of cumulus cells corresponding to each pair of bounding box images. Based on the characteristic similarity of each pair of cumulus cells, the number of cumulus cell pairs in which the two cumulus cells are the same cumulus cell is obtained.

[0012] Preferably, a deep learning classification model is used to classify cumulus cells and obtain the number of cumulus cell pairs in which two cumulus cells are the same cumulus cell.

[0013] Preferably, obtaining the training set for the deep learning classification model includes: Multiple oocytes were obtained, and two focused images were acquired from each oocyte, from which cumulus cells were extracted. Pair the cumulus cells in two focused images to form multiple pairs of cumulus cell images; Using the majority rule, based on the type of each cumulus cell in the image of each pair of cumulus cells identified by multiple people, the classification result label of each cumulus cell is obtained; All images of cumulus cells and their corresponding classification labels are obtained as the training set.

[0014] Preferably, the quality of cumulus cell removal of the oocyte to be evaluated is obtained based on the number of cumulus cells surrounding the oocyte to be evaluated and a preset quality grade, including: If the number of cumulus cells surrounding the oocyte to be evaluated is less than the first preset number, then the cumulus removal quality of the oocyte to be evaluated is excellent. If the number of cumulus cells around the oocyte to be evaluated is not less than the first preset number but less than the second preset number, then the quality of cumulus cell removal of the oocyte to be evaluated is moderate, and cumulus cell removal is performed again. If the number of cumulus cells around the oocyte to be evaluated is not less than the second preset number, then the cumulus removal quality of the oocyte to be evaluated is poor, and the oocyte to be evaluated is discarded. The first preset number is less than the second preset number.

[0015] This embodiment provides a device for assessing the quality of cumulus cell removal, including: The cumulus cell count acquisition module is used to acquire oocytes after cumulus cell removal as oocytes to be evaluated; two focused images of the oocytes to be evaluated are acquired at different focusing levels, and the location of all cumulus cells and the total number of cumulus cells in each focused image are identified; the total number of cumulus cells in the first focused image and the total number of cumulus cells in the second focused image are summed to obtain the initial count; The cumulus cell pair number acquisition module is used to extract each cumulus cell from two focused images, and pair each cumulus cell in the first focused image with each cumulus cell in the second focused image to obtain multiple cumulus cell pairs; it also acquires the number of cumulus cell pairs in which the two cumulus cells are the same cumulus cell, and based on the position of each cumulus cell, it removes cumulus cell pairs whose positions differ by a preset distance to obtain the possible number of cell pairs; The peeling quality assessment module is used to calculate the difference between the initial number and the possible number of cell pairs to obtain the number of cumulus cells around the oocyte to be assessed; based on the number of cumulus cells around the oocyte to be assessed and the preset quality level, the peeling quality of the cumulus cells of the oocyte to be assessed is obtained.

[0016] Compared with the prior art, the above-described technical solution of the present invention has the following advantages:

[0017] The cumulus cell removal quality assessment method described in this invention achieves more comprehensive coverage of the oocyte surface to be assessed by acquiring two focused images at different focal lengths, unaffected by changes in microscope field of view depth or image quality. It also sums the number of cumulus cells in the two focused images to avoid omissions due to cell positional shifts. By pairing cumulus cells in the two focused images, it identifies and removes duplicated cumulus cells. Simultaneously, based on the position between cumulus cells in a pair, it removes cumulus cell pairs with a distance greater than a preset distance, compressing the error of cross-plane repeated detection from the theoretical maximum to the actual number of matches. This yields a more accurate number of possible cell pairs, thereby quantifying the cumulus cell removal quality and achieving an objective and accurate assessment of cumulus cell removal quality.

[0018] This invention utilizes a deep learning detection model, introducing a Res2Net-COT residual structure to identify long-range dependencies between cumulus cells and capture their spatial correlations. This allows the network to acquire more contextual information and more distinctive and useful cumulus cell features, thereby improving the network's focus on cumulus cell features. Furthermore, a bidirectional feature pyramid structure is introduced, enabling the network to perform feature fusion from top to bottom and bottom to top, thus enhancing its ability to fuse cumulus cell features and avoiding the loss of detailed cumulus cell information caused by single feature fusion. High-resolution features preserve cumulus cell edge information, while low-resolution features enhance global perception of cell clusters, further improving the accuracy of the deep learning detection model in detecting cumulus cells in focused images, thereby obtaining the location and number of cumulus cells.

[0019] This invention utilizes a deep learning classification model and introduces a spatial pyramid pooling layer structure, enabling the network to convert cumulus cell feature maps of arbitrary sizes into fixed-dimensional feature vectors. This ensures that the network is not limited by the size of the input cumulus cell image and can be effectively used in conjunction with a deep learning detection model. Simultaneously, the introduction of a feature band structure allows the network to map input data to a high-dimensional feature space, thereby helping to reduce overfitting. Since the feature band structure mainly relies on local feature vectors, it can reduce overall image noise to a certain extent, improve the stability of the deep learning classification model, and further enhance the accuracy of counting possible cell pairs. Attached Figure Description

[0020] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein: Figure 1 This is a flowchart of the steps in the method for assessing the quality of cumulus cell removal according to the present invention; Figure 2 This is a schematic diagram of two focused images of the egg. Figure 3 This is a schematic diagram of cumulus cell pairing; Figure 4 This is a diagram of the reasoning process of a deep learning model. Detailed Implementation

[0021] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0022] Reference Figure 1 The flowchart shown illustrates the steps of the cumulus cell removal quality assessment method of the present invention, which include: S101: Obtain oocytes after cumulus cell removal as oocytes for evaluation; S102: At different focusing levels, two focused images of the oocyte to be evaluated are acquired, and the location of all cumulus cells and the total number of cumulus cells in each focused image are identified; S103: Sum the total number of cumulus cells in the first focused image and the total number of cumulus cells in the second focused image to obtain the initial number; S104: Extract each cumulus cell from two focused images, and pair each cumulus cell in the first focused image with each cumulus cell in the second focused image to obtain multiple cumulus cell pairs; S105: Obtain the number of cumulus cell pairs in which two cumulus cells are the same cumulus cell, and based on the position of each cumulus cell, remove cumulus cell pairs whose positions differ by a preset distance to obtain the possible number of cell pairs; S106: Calculate the difference between the initial number and the possible number of cell pairs to obtain the number of cumulus cells surrounding the oocyte to be evaluated; S107: Based on the number of cumulus cells surrounding the oocyte to be evaluated and the preset quality grade, obtain the quality of cumulus cell removal of the oocyte to be evaluated.

[0023] In this embodiment, a first focused image of the oocyte to be evaluated is acquired using a microscope at a first focusing depth; the focusing depth is then adjusted to acquire a second focused image of the oocyte to be evaluated at a second focusing depth.

[0024] The cumulus cell removal quality assessment method described in this invention achieves more comprehensive coverage of the oocyte surface to be assessed by acquiring two focused images at different focal lengths, unaffected by changes in microscope field of view depth or image quality. It also sums the number of cumulus cells in the two focused images to avoid omissions due to cell positional shifts. By pairing cumulus cells in the two focused images, it identifies and removes duplicated cumulus cells. Simultaneously, based on the position between cumulus cells in a pair, it removes cumulus cell pairs with a distance greater than a preset distance, compressing the error of cross-plane repeated detection from the theoretical maximum to the actual number of matches. This yields a more accurate number of possible cell pairs, thereby quantifying the cumulus cell removal quality and achieving an objective and accurate assessment of cumulus cell removal quality.

[0025] Based on the above embodiments, in this embodiment of the invention, a deep learning detection model is constructed based on the YOLO model. This deep learning detection model is used to identify the location of all cumulus cells and the total number of cumulus cells in each focused image, including: By introducing the Res2Net-COT residual structure into the backbone network of the YOLO model, multi-scale feature extraction is performed on the input focused image to obtain multi-scale cell morphology feature maps. A bidirectional feature pyramid structure is introduced into the neck network of the YOLO model to perform top-down and bottom-up feature fusion on the input multi-scale cell morphology feature map to obtain the fused feature map. The fused feature map is input into the detector head of the YOLO model for prediction, resulting in the bounding box for each cumulus cell; Based on the bounding boxes of all cumulus cells, the total number of cumulus cells and the location of each cumulus cell are obtained.

[0026] This invention utilizes a deep learning detection model to obtain the location and total number of all cumulus cells in a focused image. The acquisition of the training set for the deep learning detection model during training includes: Multiple oocytes were obtained, and two focused images were acquired for each oocyte; The number of focused images is increased by using image enhancement methods including rotation, horizontal flipping, vertical flipping, and contrast adjustment. All focused images are resized to a preset pixel size and manually labeled to obtain bounding boxes indicating the location of each cumulus cell in each focused image. The focused images and their corresponding bounding boxes are used as the training set.

[0027] This invention utilizes a deep learning detection model, introducing a Res2Net-COT residual structure to identify long-range dependencies between cumulus cells and capture their spatial correlations. This allows the network to acquire more contextual information and more distinctive and useful cumulus cell features, thereby improving the network's focus on cumulus cell features. Furthermore, a bidirectional feature pyramid structure is introduced, enabling the network to perform feature fusion from top to bottom and bottom to top, thus enhancing its ability to fuse cumulus cell features and avoiding the loss of detailed cumulus cell information caused by single feature fusion. High-resolution features preserve cumulus cell edge information, while low-resolution features enhance global perception of cell clusters, further improving the accuracy of the deep learning detection model in detecting cumulus cells in focused images, thereby obtaining the location and number of cumulus cells.

[0028] Based on the above embodiments, in this embodiment of the invention, a deep learning classification model is constructed based on Siamese or SwAV, and the deep learning classification model is used to obtain the number of cumulus cell pairs in which two cumulus cells are the same cumulus cell, including: A spatial pyramid pooling layer structure is introduced into the feature extraction network of the model. The bounding box images of cumulus cells in two focused images are paired as input to extract the features of cumulus cells in each bounding box image and convert them into feature vectors of fixed dimensions. A feature band structure is introduced into the feature contrast network of the model to map the feature vectors to a high-dimensional feature space, obtain the corresponding denoised feature map, and calculate the contrast loss function value between each pair of bounding box images as the feature similarity of a pair of cumulus cells corresponding to each pair of bounding box images. Based on the characteristic similarity of each pair of cumulus cells, the number of cumulus cell pairs in which the two cumulus cells are the same cumulus cell is obtained.

[0029] This invention utilizes a deep learning classification model to classify cumulus cells and obtain the number of cumulus cell pairs where both cumulus cells belong to the same cumulus cell. When training the deep learning classification model, the acquisition of the training set includes: Multiple oocytes were obtained, and two focused images were acquired from each oocyte, from which cumulus cells were extracted. Pair the cumulus cells in two focused images to form multiple pairs of cumulus cell images; Using the majority rule, based on the type of each cumulus cell in the image of each pair of cumulus cells identified by multiple people, the classification result label of each cumulus cell is obtained; All images of cumulus cells and their corresponding classification labels are obtained as the training set.

[0030] This invention utilizes a deep learning classification model and introduces a spatial pyramid pooling layer structure, enabling the network to convert cumulus cell feature maps of arbitrary sizes into fixed-dimensional feature vectors. This ensures that the network is not limited by the size of the input cumulus cell image and can be effectively used in conjunction with a deep learning detection model. Simultaneously, the introduction of a feature band structure allows the network to map input data to a high-dimensional feature space, thereby helping to reduce overfitting. Since the feature band structure mainly relies on local feature vectors, it can reduce overall image noise to a certain extent, improve the stability of the deep learning classification model, and further enhance the accuracy of counting possible cell pairs.

[0031] Specifically, based on the number of cumulus cells surrounding the oocyte to be evaluated and a preset quality grade, the quality of cumulus cell removal of the oocyte to be evaluated is obtained, including: If the number of cumulus cells surrounding the oocyte to be evaluated is less than the first preset number, then the cumulus removal quality of the oocyte to be evaluated is excellent. If the number of cumulus cells around the oocyte to be evaluated is not less than the first preset number but less than the second preset number, then the quality of cumulus cell removal of the oocyte to be evaluated is moderate, and cumulus cell removal is performed again. If the number of cumulus cells around the oocyte to be evaluated is not less than the second preset number, then the cumulus removal quality of the oocyte to be evaluated is poor, and the oocyte to be evaluated is discarded. The first preset number is less than the second preset number.

[0032] This invention provides a method for assessing the quality of cumulus cell removal. This method can automatically obtain the number of all cumulus cells around the oocyte from two focused images using deep learning technology. It is not limited by the depth of the microscope field of view or changes in image quality, and can achieve objective, accurate and automatic assessment of the quality of cumulus cell removal, providing more assistance for ART operations.

[0033] Based on the above embodiments, in this embodiment of the invention, the method provided by the present invention is used to assess the quality of cumulus cell removal, including: S201: Input the two focused images of the oocyte into the trained deep learning detection model to obtain the location and number of cumulus cells in the two focused images respectively. , The total number of cumulus cells was obtained from two focused images. ; S202: Based on the location of cumulus cells in the focused image, extract cumulus cells from two separate focused images, pair them up, and input them into a pre-trained deep learning classification model to classify the cumulus cells and obtain the number of pairs of identical cumulus cells. ; S203: Eliminated based on the identified cumulus cell locations. For logarithms with large differences in center position, we get For cumulus cells; S204: Number of all cumulus cells surrounding the exported oocyte. After the cumulus cell removal procedure, if If the removal of the cumulus cells is complete, then the removal of the cumulus cells is considered complete; otherwise, the removal of the cumulus cells is considered incomplete.

[0034] Compared to existing technologies, the cumulus cell removal quality assessment method of this invention automatically obtains the number of all cumulus cells surrounding the oocyte using two focused images, overcoming the limitations of microscope field of view depth and physician subjectivity. Simultaneously, this method can learn useful features within a certain range of noise and brightness variations, thereby reducing the impact of image noise and brightness changes on the accuracy of the removal quality assessment. Therefore, this method enables objective, accurate, and automated assessment of cumulus cell removal quality, thus improving the success rate of ART procedures.

[0035] Reference Figure 2 The image shown is a schematic diagram of two focused images of an egg; specifically, the process of creating the dataset for the deep learning detection model includes: S301: Prepare 200 oocytes, and repeatedly aspirate and expel them using a microtube to acquire two focused images for each oocyte, with image depths of 0μm and -10μm respectively; To increase the amount of data, methods such as rotation, horizontal flipping, vertical flipping, and contrast adjustment were used to expand the number of images to 2000; 1000 images each with a depth of 0μm and -10μm were also used.

[0036] S302: Resize the 2000 images to 224 pixels × 224 pixels to fit the input size of the detection model. Simultaneously, manually annotate these 2000 images using the LabelImg annotation software to generate corresponding label data. The label data format can be converted to *.txt format using code.

[0037] S303: Based on a 6:2:2 cutting ratio, divide these 2000 images and their corresponding labels into the training set, validation set, and test set for the model.

[0038] Reference Figure 3 The diagram shown illustrates cumulus cell pairing. The dataset creation process for the deep learning classification model of this invention includes the following steps: S401: Based on the above 2000 oocyte images (1000 each of 0μm and -10μm), manual cropping was performed to extract cumulus cells from each COC image, and these cumulus cells were paired up to obtain 25000 pairs of cumulus cell images.

[0039] S402: Four experienced operators evaluate the cumulus cells by voting and determine the classification result label (same or different) of the cumulus cells according to the majority rule.

[0040] S403: Obtain 25,000 pairs of images and their corresponding labels, and divide them into training, validation and test sets for the model according to a cutting ratio of 8:1:1.

[0041] Reference Figure 4 The diagram shown illustrates the inference process of a deep learning model. The inference process of the deep learning model of this invention includes the following steps: S501: The focused image is processed and cumulus cell features are extracted by the backbone convolutional network (CBL+CSP1) in the deep learning detection model I. These features are used to infer the location of cumulus cells in the form of rectangular prediction boxes in the last layer of the network and are then cropped. S502: The extracted cumulus cells are paired and input into the deep learning classification model II. The model first extracts the feature vectors of the paired cumulus cells through a backbone convolutional network (CNN), then compares the feature similarity of the paired cumulus cells through a contrastive loss function, and finally infers the category (same or different) of the paired cumulus cells.

[0042] This embodiment, based on the existing deep learning detection model (YOLO), introduces a residual (Res2Net-COT) structure into the backbone network of the model, enabling the network to acquire more contextual information and more distinctive and useful cumulus cell features, thereby improving the network's focus on cumulus cell features. Based on the existing deep learning detection model (YOLO), a bidirectional feature pyramid (BiFPN) structure is introduced into the neck network of the model, enabling the network to perform feature fusion from top to bottom and from bottom to top, thereby improving the network's ability to fuse cumulus cell features.

[0043] This embodiment, based on the existing deep learning classification model (Siamese), introduces a Spatial Pyramid Pooling Layer (SPP) structure into the model's feature extraction network. This allows the network to convert cumulus cell feature maps of arbitrary sizes into fixed-dimensional feature vectors, ensuring that the network is not limited by the size of the input cumulus cell image and can be effectively used in conjunction with deep learning detection models. Based on the existing deep learning classification model (Siamese), a Feature Band of Functions (BoF) structure is introduced into the model's feature contrast network, enabling the network to map input data to a high-dimensional feature space, thereby helping to reduce overfitting. Since the BoF structure mainly relies on local feature vectors, it can reduce overall image noise to a certain extent and improve the stability of the network model.

[0044] To further test the performance of the deep learning detection model of this invention, 400 proportionally segmented oocyte images (200 each of 0μm and -10μm) and their corresponding labels were used as the test set for model performance testing. The results show that the deep learning detection model of this embodiment has significant inference performance, with a mean average precision (mAP) of 92.3%, a recall of 85.4%, and a detection speed (FPS) of 26.1, as detailed in Table 1. Table 1: Performance of the cumulus cell detection model performance mAP / (%) Recall / (%) FPS (frames per second) Detection model 92.3 85.4 26.1

[0045] To further test the performance of the deep learning classification model of this invention, 400 proportionally segmented oocyte images (200 each of 0μm and -10μm) and their corresponding labels were used as the test set for model performance testing. The results show that the deep learning classification model of this invention has significant inference performance, with an accuracy of 94.6%, an F1 score of 92.2%, and an area under the ROC curve (AUC) of 91.0%, as detailed in Table 2. Table 2: Performance of the cumulus cell classification model performance Accuracy / (%) F1 score / (%) AUC / (%) Classification model 94.6 92.2 91.0

[0046] To further test the accuracy of the cumulus cell removal quality assessment method of the present invention, 150 oocytes were divided into three groups: a manual observation group, an image thresholding group, and a deep learning model group. During the cumulus cell removal operation, the methods of these three groups were used to infer the cumulus cell removal quality. The results showed that the deep learning model method of the present invention had the highest inference accuracy, reaching 98.3%, while the image thresholding method and the manual observation method only achieved accuracy of 90.5% and 82.4%, respectively. See Table 3 for details. Table 3: Comparison of the accuracy of different methods for assessing the quality of cumulus cell removal performance Accuracy / (%) manual observation 82.4% Image threshold 90.5% Deep learning models 98.3%

[0047] This invention improves the accuracy of deep learning detection models in detecting cumulus cells in focused images by introducing Res2Net-COT and BiFPN structures, thereby obtaining the location and number of cumulus cells; and improves the accuracy of deep learning classification models in classifying cumulus cells in focused images by introducing SPP and BoF structures, thereby obtaining the same cumulus cells and different cumulus cells.

[0048] Based on the above embodiments, this invention also provides a device for assessing the quality of cumulus cell removal; the specific device may include: The cumulus cell count acquisition module is used to acquire oocytes after cumulus cell removal as oocytes to be evaluated; two focused images of the oocytes to be evaluated are acquired at different focusing levels, and the location of all cumulus cells and the total number of cumulus cells in each focused image are identified; the total number of cumulus cells in the first focused image and the total number of cumulus cells in the second focused image are summed to obtain the initial count; The cumulus cell pair number acquisition module is used to extract each cumulus cell from two focused images, and pair each cumulus cell in the first focused image with each cumulus cell in the second focused image to obtain multiple cumulus cell pairs; it also acquires the number of cumulus cell pairs in which the two cumulus cells are the same cumulus cell, and based on the position of each cumulus cell, it removes cumulus cell pairs whose positions differ by a preset distance to obtain the possible number of cell pairs; The peeling quality assessment module is used to calculate the difference between the initial number and the possible number of cell pairs to obtain the number of cumulus cells around the oocyte to be assessed; based on the number of cumulus cells around the oocyte to be assessed and the preset quality level, the peeling quality of the cumulus cells of the oocyte to be assessed is obtained.

[0049] The cumulus cell removal quality assessment device of this embodiment is used to implement the aforementioned cumulus cell removal quality assessment method. Therefore, the specific implementation of the cumulus cell removal quality assessment device can be found in the embodiment section of the cumulus cell removal quality assessment method above. For example, the cumulus cell number acquisition module is used to implement steps S101, S102, and S103 in the aforementioned cumulus cell removal quality assessment method; the cumulus cell pair number acquisition module is used to implement steps S104 and S105 in the aforementioned cumulus cell removal quality assessment method; and the removal quality assessment module is used to implement steps S106 and S107 in the aforementioned cumulus cell removal quality assessment method. Therefore, its specific implementation can be referred to the description of the corresponding embodiments, and will not be repeated here.

[0050] The cumulus cell removal quality assessment method described in this invention achieves more comprehensive coverage of the oocyte surface to be assessed by acquiring two focused images at different focal lengths, unaffected by changes in microscope field of view depth or image quality. It also sums the number of cumulus cells in the two focused images to avoid omissions due to cell positional shifts. By pairing cumulus cells in the two focused images, it identifies and removes duplicated cumulus cells. Simultaneously, based on the position between cumulus cells in a pair, it removes cumulus cell pairs with a distance greater than a preset distance, compressing the error of cross-plane repeated detection from the theoretical maximum to the actual number of matches. This yields a more accurate number of possible cell pairs, thereby quantifying the cumulus cell removal quality and achieving an objective and accurate assessment of cumulus cell removal quality. This invention utilizes a deep learning detection model, introducing a Res2Net-COT residual structure to identify long-range dependencies between cumulus cells and capture their spatial correlations. This allows the network to acquire more contextual information and more distinctive and useful cumulus cell features, thereby improving the network's focus on cumulus cell features. Furthermore, a bidirectional feature pyramid structure is introduced, enabling the network to perform feature fusion from top to bottom and bottom to top, thus enhancing its ability to fuse cumulus cell features and avoiding the loss of detailed cumulus cell information caused by single feature fusion. High-resolution features preserve cumulus cell edge information, while low-resolution features enhance global perception of cell clusters, further improving the accuracy of the deep learning detection model in detecting cumulus cells in focused images, thereby obtaining the location and number of cumulus cells. This invention utilizes a deep learning classification model and introduces a spatial pyramid pooling layer structure, enabling the network to convert cumulus cell feature maps of arbitrary sizes into fixed-dimensional feature vectors. This ensures that the network is not limited by the size of the input cumulus cell image and can be effectively used in conjunction with a deep learning detection model. Simultaneously, the introduction of a feature band structure allows the network to map input data to a high-dimensional feature space, thereby helping to reduce overfitting. Since the feature band structure mainly relies on local feature vectors, it can reduce overall image noise to a certain extent, improve the stability of the deep learning classification model, and further enhance the accuracy of counting possible cell pairs.

[0051] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0052] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0053] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0054] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0055] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for assessing the quality of cumulus cell removal, characterized in that, include: Oocytes obtained after cumulus cell removal were used as oocytes for evaluation; Two focused images of the oocyte to be evaluated were acquired at different focusing levels, and the location of all cumulus cells and the total number of cumulus cells in each focused image were identified. The initial number is obtained by summing the total number of cumulus cells in the first focused image and the total number of cumulus cells in the second focused image; Each cumulus cell in two focused images is extracted, and each cumulus cell in the first focused image is paired with each cumulus cell in the second focused image to obtain multiple cumulus cell pairs; Obtain the number of cumulus cell pairs in which the two cumulus cells are the same cumulus cell, and based on the position of each cumulus cell, remove cumulus cell pairs whose positions differ by a preset distance to obtain the possible number of cell pairs; The difference between the initial number and the possible number of cell pairs is calculated to obtain the number of cumulus cells surrounding the oocyte to be evaluated; The quality of cumulus cell removal of the oocyte is obtained based on the number of cumulus cells surrounding the oocyte to be evaluated and the preset quality grade.

2. The method for assessing the quality of cumulus cell removal according to claim 1, characterized in that, Two focused images of the oocyte to be evaluated after cumulus cell removal were obtained, including: The first-focus image of the oocyte to be evaluated was acquired using a microscope at the first focal depth. Adjust the depth of focus and acquire a second focused image of the oocyte to be evaluated at the second depth of focus.

3. The method for assessing the quality of cumulus cell removal according to claim 1, characterized in that, A deep learning detection model was built based on the YOLO model. This model was used to identify the location and total number of all cumulus cells in each focused image, including: By introducing the Res2Net-COT residual structure into the backbone network of the YOLO model, multi-scale feature extraction is performed on the input focused image to obtain multi-scale cell morphology feature maps. A bidirectional feature pyramid structure is introduced into the neck network of the YOLO model to perform top-down and bottom-up feature fusion on the input multi-scale cell morphology feature map to obtain the fused feature map. The fused feature map is input into the detector head of the YOLO model for prediction, resulting in the bounding box for each cumulus cell; Based on the bounding boxes of all cumulus cells, the total number of cumulus cells and the location of each cumulus cell are obtained.

4. The method for assessing the quality of cumulus cell removal according to claim 1, characterized in that, Using a deep learning detection model, the location of all cumulus cells and the total number of cumulus cells in the focused image are obtained.

5. The method for assessing the quality of cumulus cell removal according to claim 4, characterized in that, Obtaining the training set for a deep learning detection model includes: Multiple oocytes were obtained, and two focused images were acquired for each oocyte; The number of focused images is increased by using image enhancement methods including rotation, horizontal flipping, vertical flipping, and contrast adjustment. All focused images are resized to a preset pixel size and manually labeled to obtain bounding boxes indicating the location of each cumulus cell in each focused image. The focused images and their corresponding bounding boxes are used as the training set.

6. The method for assessing the quality of cumulus cell removal according to claim 1, characterized in that... A deep learning classification model is built based on Siamese or SwAV. This model is then used to obtain the number of cumulus cell pairs in which two cumulus cells belong to the same cumulus cell. This includes: A spatial pyramid pooling layer structure is introduced into the feature extraction network of the model. The bounding box images of cumulus cells in two focused images are paired as input to extract the features of cumulus cells in each bounding box image and convert them into feature vectors of fixed dimensions. A feature band structure is introduced into the feature contrast network of the model to map the feature vectors to a high-dimensional feature space, obtain the corresponding denoised feature map, and calculate the contrast loss function value between each pair of bounding box images as the feature similarity of a pair of cumulus cells corresponding to each pair of bounding box images. Based on the characteristic similarity of each pair of cumulus cells, the number of cumulus cell pairs in which the two cumulus cells are the same cumulus cell is obtained.

7. The method for assessing the quality of cumulus cell removal according to claim 1, characterized in that, Using a deep learning classification model, cumulus cells are classified, and the number of cumulus cell pairs in which two cumulus cells belong to the same cumulus cell is obtained.

8. The method for assessing the quality of cumulus cell removal according to claim 7, characterized in that, Obtaining the training set for a deep learning classification model includes: Multiple oocytes were obtained, and two focused images were acquired from each oocyte, from which cumulus cells were extracted. Pair the cumulus cells in two focused images to form multiple pairs of cumulus cell images; Using the majority rule, based on the type of each cumulus cell in the image of each pair of cumulus cells identified by multiple people, the classification result label of each cumulus cell is obtained; All images of cumulus cells and their corresponding classification labels are obtained as the training set.

9. The method for assessing the quality of cumulus cell removal according to claim 1, characterized in that, Based on the number of cumulus cells surrounding the oocyte to be evaluated and the preset quality grade, the quality of cumulus cell removal of the oocyte to be evaluated is obtained, including: If the number of cumulus cells surrounding the oocyte to be evaluated is less than the first preset number, then the cumulus removal quality of the oocyte to be evaluated is excellent. If the number of cumulus cells around the oocyte to be evaluated is not less than the first preset number but less than the second preset number, then the quality of cumulus cell removal of the oocyte to be evaluated is moderate, and cumulus cell removal is performed again. If the number of cumulus cells around the oocyte to be evaluated is not less than the second preset number, then the cumulus removal quality of the oocyte to be evaluated is poor, and the oocyte to be evaluated is discarded. The first preset number is less than the second preset number.

10. A device for assessing the quality of cumulus cell removal, characterized in that, include: The cumulus cell number acquisition module is used to obtain oocytes after cumulus cell removal as oocytes to be evaluated; Two focused images of the oocyte to be evaluated were acquired at different focusing levels, and the location of all cumulus cells and the total number of cumulus cells in each focused image were identified. The initial number is obtained by summing the total number of cumulus cells in the first focused image and the total number of cumulus cells in the second focused image; The cumulus cell pair number acquisition module is used to extract each cumulus cell from two focused images, and pair each cumulus cell in the first focused image with each cumulus cell in the second focused image to obtain multiple cumulus cell pairs; it also acquires the number of cumulus cell pairs in which the two cumulus cells are the same cumulus cell, and based on the position of each cumulus cell, it removes cumulus cell pairs whose positions differ by a preset distance to obtain the possible number of cell pairs; The peeling quality assessment module is used to calculate the difference between the initial number and the possible number of cell pairs to obtain the number of cumulus cells around the oocyte to be assessed; based on the number of cumulus cells around the oocyte to be assessed and the preset quality level, the peeling quality of the cumulus cells of the oocyte to be assessed is obtained.

Images