A method and system for evaluating oocyte quality

Abstract

The application relates to the field of biological medical technology, and specifically provides an oocyte quality evaluation method and system, which comprises the following steps: inputting an oocyte image into an image segmentation model, obtaining coding feature maps of different scales through a plurality of residual convolution blocks, and generating a final coding feature map; extracting multi-scale context features of the final coding feature map in parallel through a plurality of hollow convolution kernels, and performing splicing and fusion to obtain a fusion feature map; splicing and fusing the up-sampling decoding result of the fusion feature map and the coding feature map through a jump connection to obtain a final decoding feature map; outputting a segmentation mask image through a multi-channel output layer; obtaining morphological feature parameters of the oocyte through a contour fitting image algorithm, inputting the morphological feature parameters into a preset oocyte scoring model, and obtaining an oocyte quality evaluation result, so as to solve the problem that the image structure segmentation of the oocyte is not accurate in the related art, and the oocyte quality evaluation accuracy is affected.

Description

Technical Field

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

[0002] With the accelerated pace of modern life, increased environmental pressure, and a clear trend towards later marriage and childbearing, the incidence of infertility among the global reproductive-age population is gradually rising. Assisted reproductive technologies such as in vitro fertilization (IVF) and intracytoplasmic sperm injection (ICSI) are increasingly becoming effective treatments for infertility. Oocyte quality is a crucial factor affecting the success rate of assisted reproductive procedures. Oocyte quality assessment involves embryologists observing the morphological characteristics of oocytes under an inverted microscope, including parameters such as oocyte diameter, polar body morphology (ovality and area), and perivitelline space. However, this method suffers from high subjectivity, poor repeatability, and limited quantitative capabilities. Inaccurate oocyte quality assessment can lead to decreased fertilization rates and hindered embryo development during subsequent IVF or embryo culture stages, thus affecting the success rate of assisted reproductive procedures. Therefore, objective, accurate, and automated assessment of oocyte quality is of great significance for improving the success rate of assisted reproductive procedures and enhancing clinical outcomes.

[0003] Currently, oocyte quality assessment mainly relies on embryologists' manual observation and subjective judgment of morphological characteristics such as oocyte diameter, polar body morphology, and perivitelline space under an inverted microscope. This method depends on the operator's clinical experience and subjective judgment, resulting in significant differences in assessment results among different embryologists. Furthermore, the manual assessment process is time-consuming, labor-intensive, and lacks repeatability, making it difficult to meet the growing clinical needs of assisted reproduction. In addition, some studies use image thresholding to automatically segment oocyte microscopic images to extract morphological feature parameters. However, due to the small grayscale differences and blurred boundaries between oocytes, zona pellucida, and polar bodies in oocyte microscopic images, image thresholding is sensitive to uneven lighting and noise interference, easily leading to inaccurate structural segmentation and incomplete boundary identification, resulting in large errors in the measurement of morphological feature parameters. Moreover, most existing automatic assessment methods only analyze single morphological indicators, lacking the ability to comprehensively evaluate multiple morphological characteristics of oocytes, making it difficult to fully reflect the quality status and developmental potential of oocytes.

[0004] There is currently no effective solution to the problem of inaccurate image structure segmentation of oocytes in related technologies, which affects the accuracy of oocyte quality assessment. Summary of the Invention

[0005] The present invention provides a method and system for assessing oocyte quality, which at least solves the problem in related technologies where inaccurate image structure segmentation of oocytes affects the accuracy of oocyte quality assessment.

[0006] According to one aspect of the present invention, a method for assessing oocyte quality is provided, comprising: inputting an oocyte image into an image segmentation model, sequentially performing downsampling encoding processing through multiple residual convolutional blocks to obtain encoded feature maps at different scales, and generating a final encoded feature map at the deepest layer; extracting multi-scale contextual features of the final encoded feature map in parallel using multiple dilated convolutional kernels with different dilation rates, and concatenating and fusing the multi-scale contextual features to obtain a fused feature map; concatenating and fusing the upsampling decoding result of the fused feature map with the encoded feature map at the corresponding level in the downsampling encoding through skip connections to obtain a final decoded feature map; performing pixel classification on the final decoded feature map through a multi-channel output layer to output a segmentation mask map; wherein the multi-channel output layer outputs segmentation mask maps corresponding to multiple morphological features of the oocyte; based on the segmentation mask map, obtaining morphological feature parameters of the oocyte through a contour fitting image algorithm, inputting the morphological feature parameters into a preset oocyte scoring model to obtain an oocyte quality assessment result.

[0007] As an optional approach, the morphological characteristic parameters are input into a preset oocyte scoring model to obtain the quality assessment result of the oocyte, including: determining the assessment indicators of the oocyte based on the morphological characteristic parameters; wherein the assessment indicators include oocyte diameter, polar body ellipticity, polar body area, and perivitelline space; obtaining the indicator assessment score corresponding to the assessment indicator based on the scoring criteria corresponding to the assessment indicator; inputting the indicator assessment score into the preset oocyte scoring model to obtain a comprehensive assessment score; and comparing the comprehensive assessment score with a preset threshold to obtain the quality assessment result.

[0008] As an optional approach, the evaluation indicators of the oocyte are determined based on the morphological characteristic parameters, including: the expression for the polar body ellipticity is as follows: Where e represents the ellipticity of the polar body, a represents the major axis of the elliptical profile of the polar body, and b represents the minor axis of the elliptical profile of the polar body; the expression for the area of ​​the polar body is as follows: Where s represents the polar body area and p represents the pixel distance calibration value; the expression for the periovarian gap is as follows: Where d represents the periovarian space, D represents the diameter of the inner ring of the zona pellucida, and D represents the diameter of the oocyte.

[0009] As an optional approach, the segmentation mask image includes an oocyte, a polar body, and a zona pellucida; the morphological feature parameters include the oocyte diameter, the major and minor axes of the polar body, and the inner ring diameter of the zona pellucida; based on the segmentation mask image, the morphological feature parameters of the oocyte are obtained through a contour fitting image algorithm, including: obtaining the oocyte diameter based on the oocyte segmentation mask image through circular contour fitting; obtaining the major and minor axes of the elliptical contour of the polar body based on the polar body segmentation mask image through elliptical contour fitting; and obtaining the inner ring diameter of the zona pellucida based on the zona pellucida segmentation mask image through annular contour fitting.

[0010] As an optional approach, the expression for the oocyte scoring model is as follows: ;in, The index score representing the diameter of the oocyte. The index score representing the ellipticity of the polar body. This represents the evaluation score corresponding to the polar body area. This indicates the assessment score of the index corresponding to the periovarian space.

[0011] As an optional approach, the evaluation score corresponding to the evaluation index is obtained according to the scoring criteria corresponding to the evaluation index, including: setting a benchmark range according to the quality evaluation requirements of the oocyte; determining that the evaluation score corresponding to the evaluation index is full marks when the evaluation index is within the benchmark range; and subtracting a preset score from the full marks when the evaluation index exceeds the benchmark range based on the deviation between the evaluation index and the benchmark range. The preset score is proportional to the absolute value of the deviation, and the preset score is greater than zero and less than or equal to the full marks.

[0012] As an optional approach, before inputting the oocyte image into the image segmentation model, the method further includes: acquiring multiple oocyte image samples, wherein the image samples include the oocyte, polar body, and zona pellucida; performing geometric transformation and image enhancement on the oocyte image samples to obtain enhanced image samples; wherein the geometric transformation and image enhancement include at least one of rotation, horizontal flipping, vertical flipping, and contrast adjustment; manually labeling the enhanced image samples to generate corresponding labels; dividing the enhanced image samples and the corresponding labels into a training set, a validation set, and a test set according to a preset ratio; and training the image segmentation model based on the training set, validation set, and test set to obtain a trained image segmentation model.

[0013] According to another aspect of the present invention, an oocyte quality assessment system is also provided for performing the above-described oocyte quality assessment method, comprising: an acquisition module, a segmentation module, and an assessment module; the acquisition module is used to acquire oocyte images and input the oocyte images into the segmentation module; the segmentation module is connected to the acquisition module and is used to input the oocyte images into an image segmentation model, sequentially perform downsampling encoding processing through multiple residual convolutional blocks to obtain encoded feature maps of different scales, and generate a final encoded feature map in the deepest layer; multi-scale contextual features of the final encoded feature map are extracted in parallel through multiple dilated convolutional kernels with different dilation rates, and the multi-scale contextual features are concatenated. The fusion process yields a fused feature map. The upsampled decoding result of the fused feature map is then combined with the corresponding level of the downsampled encoding feature map via skip connections to obtain the final decoded feature map. A multi-channel output layer performs pixel classification on the final decoded feature map, outputting a segmentation mask map. The multi-channel output layer outputs segmentation masks representing various morphological features of the oocyte. The evaluation module, connected to the segmentation module, uses the segmentation mask map and a contour fitting image algorithm to obtain the morphological feature parameters of the oocyte. These morphological feature parameters are then input into a preset oocyte scoring model to obtain the oocyte quality evaluation result.

[0014] According to another aspect of the present invention, an electronic device is also provided, comprising: a processor, and a memory storing a program, the program including instructions that, when executed by the processor, cause the processor to perform the method according to any one of the preceding claims.

[0015] According to another aspect of the invention, a non-transitory machine-readable medium storing computer instructions for causing the computer to perform any of the methods described above is also provided.

[0016] The present invention provides an oocyte quality assessment method and system that employs an image segmentation model and downsamples and encodes multiple residual convolutional blocks to extract multi-level semantic features from shallow to deep layers in oocyte images, effectively addressing the problem of small grayscale differences. By extracting multi-scale contextual features in parallel using multiple dilated convolutional kernels with different dilation rates and then splicing and fusing them, a larger receptive field can be obtained without sacrificing spatial resolution. This allows the model to fully perceive the global positional relationships and contextual information between various structures of the oocyte, thereby accurately distinguishing structural regions with similar grayscale but different morphologies.

[0017] Simultaneously, by using skip connections to fuse the encoded feature maps from the downsampling encoding stage with the upsampling decoding results, the loss of spatial information during downsampling is compensated for, resulting in more refined and complete segmentation boundaries. Finally, a multi-channel output layer outputs segmentation mask maps of various morphological features, achieving simultaneous and accurate segmentation of multiple morphological features. The segmentation mask maps are then used to calculate morphological feature parameters through a contour fitting algorithm, which are then input into a scoring model to obtain objective and accurate oocyte quality assessment results. This solves the problem of inaccurate oocyte image structure segmentation in related technologies, which affects the accuracy of oocyte quality assessment. Attached Figure Description

[0018] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other embodiments based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of an oocyte quality assessment method according to an embodiment of the present invention.

[0020] Figure 2 This is a structural diagram of the image segmentation model according to an embodiment of the present invention.

[0021] Figure 3 This is a schematic diagram of a microscopic image of a primitive oocyte created in this invention.

[0022] Figure 4 This is a schematic diagram of the oocyte quality assessment system created in this invention.

[0023] Figure 5 This is a schematic diagram of the structure of the electronic device created by this invention. Detailed Implementation

[0024] Embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0025] To address the problem of inaccurate image structure segmentation of oocytes in related technologies, which affects the accuracy of oocyte quality assessment, this invention provides an oocyte quality assessment method, such as... Figure 1 As shown, it includes: Step S101: Input the oocyte image into the image segmentation model, and perform downsampling encoding processing through multiple residual convolution blocks in sequence to obtain encoded feature maps of different scales, and generate the final encoded feature map in the deepest layer; Step S102: Multi-scale contextual features of the final encoded feature map are extracted in parallel using multiple dilated convolutional kernels with different dilation rates, and the multi-scale contextual features are concatenated and fused to obtain a fused feature map. Step S103: The upsampled decoding result of the fused feature map after upsampling decoding is concatenated and fused with the corresponding level of the encoded feature map in the downsampled encoding through skip connections to obtain the final decoded feature map; Step S104: The final decoded feature map is classified by pixels through a multi-channel output layer to output a segmentation mask map; wherein, the multi-channel output layer corresponds to the output of segmentation mask maps of various morphological features of oocytes; Step S105: Based on the segmentation mask image, obtain the morphological feature parameters of the oocyte through the contour fitting image algorithm, input the morphological feature parameters into the preset oocyte scoring model, and obtain the quality assessment result of the oocyte.

[0026] The above steps are executed by computer equipment or a dedicated image processing system, specifically through electronic devices equipped with image segmentation models and preset oocyte scoring models. The image segmentation model, contour fitting algorithm, and preset oocyte scoring model in the method are all computer program modules that run on a processor to achieve automated evaluation. Users only need to acquire microscopic images of oocytes through image acquisition equipment and input these images into the equipment or system; all subsequent feature extraction, contour fitting, parameter calculation, and quality scoring steps are automatically executed by the computer program without manual intervention.

[0027] Image segmentation models transform two-dimensional image information into high-precision multi-channel segmentation masks, physically separating and digitally extracting different morphological features in spatial location. This pixel-level precision enables subsequent contour fitting algorithms to calculate objective, continuous, and quantifiable morphological feature parameters, providing standardized and repeatable structured data input for the pre-defined oocyte scoring model, ultimately achieving objective and accurate oocyte quality assessment.

[0028] Image segmentation models are semantic segmentation networks based on deep learning. Image segmentation models include, but are not limited to, improved UNet (U-shaped network), MaskRcnn (mask region-based convolutional neural network), SegNet (semantic segmentation network), etc. The choice of model depends on the specific task requirements, the accuracy requirements, and the available computing resources.

[0029] like Figure 2 As shown, this embodiment preferably incorporates an improved UNet segmentation network that integrates a residual structure downsampling encoder, a multi-scale feature extraction module with a dilated spatial convolution structure, and a skip connection upsampling decoder. This encoder-decoder structure performs fine-grained classification of each pixel in the image, thereby outlining the contours of the oocyte and its surrounding structures. In this embodiment, the final output layer has four channels. This number of channels ensures that the black background, oocyte, polar body, and zona pellucida are all output, preventing the loss of oocyte morphological features. Figure 2 The text shows three types of target regions that require quantitative calculation, including oocytes, polar bodies, and zona pellucida.

[0030] Residual convolutional blocks are deep convolutional neural network modules that introduce skip connections. Residual convolutional blocks add the input features element-wise to the output after processing through several convolutional layers, batch normalization, and activation functions, thereby effectively alleviating the gradient vanishing and degradation problems in deep networks, enabling the network to build deeper feature extraction layers without sacrificing training efficiency.

[0031] In this embodiment, the residual convolutional blocks progressively reduce the spatial resolution of the feature maps, enabling each convolutional block to learn larger-scale semantic information within a smaller spatial range. Specifically, shallow blocks preserve local texture details such as cell membranes and particles, while deep blocks capture global context such as contour morphology and overall structure, thus naturally generating encoded feature maps of different scales at different downsampling levels.

[0032] The deepest layer generates the final encoded feature map, which refers to the layer after passing through the largest number of stacked residual convolutional blocks and completing the final downsampling. This layer's feature map possesses the largest receptive field and the most abstract semantic expression, meaning that the feature representation with the highest semantic condensation is used as the input for subsequent multi-scale dilated convolutional fusion. This allows for the extraction of the richest global structural information while compressing spatial details, providing a deep discriminative basis for accurately segmenting oocyte morphology.

[0033] A dilated convolution kernel is an improved convolution operation that inserts holes, or zero-value intervals, between the pixels of a standard convolution kernel. Without increasing the number of parameters or computational cost, it exponentially expands the receptive field of the convolution kernel by adjusting the dilation rate.

[0034] The final encoded feature map has the highest semantic condensation but loses spatial details. Furthermore, different features in oocyte images, such as zona pellucida thickness, perivitelline space width, and polar body morphology, have vastly different scale ranges. This embodiment uses dilated convolutions with varying dilation rates in parallel, allowing the same network layer to simultaneously see fine textures at small dilation rates and large structural contexts at large dilation rates. This integrates complementary information captured from different receptive fields into a unified feature representation.

[0035] Multi-scale refers to multiple spatial scale levels covering the receptive field, ranging from local to global. In this embodiment, the final encoded feature map of the deepest, highly abstract semantics is recoupled with multi-scale spatial context features to form a fused feature map that simultaneously contains large-scale structural relationships and small-scale boundary details, serving as the input basis for subsequent upsampling decoding.

[0036] During the multi-level downsampling encoding process, the spatial resolution of the image is repeatedly compressed. Although this yields deep features rich in global semantics, it loses the precise spatial location information at the pixel level. Since the final segmentation task requires classifying each pixel of the original image, it is necessary to gradually restore the reduced feature map to the original input size through upsampling operations.

[0037] The upsampling decoding process produces an upsampling decoding result, which is the spatial feature map with a specific resolution recovered at the current level. In downsampling encoding, the corresponding level refers to the encoding stage that exactly matches the resolution of the current upsampling step. For example, when the decoder recovers 1 / 32 resolution to 1 / 16 through a single upsampling, the corresponding 1 / 16 resolution encoded feature map output by the encoder after the third downsampling step has the same spatial size.

[0038] The features recovered by the decoder itself emphasize semantic discriminative power but have blurred boundaries, while the features of the same layer in the encoder emphasize spatially precise information such as edges and textures but lack overall discriminative power. By fusing the upsampled decoding results and the corresponding layer's encoded feature maps in the downsampled encoding, both of these complementary advantages can be preserved simultaneously.

[0039] Skip connections are feature transfer channels in deep convolutional neural networks that bypass one or more intermediate layers and directly pass the output of shallow layers to the input of deep layers. In the encoder-decoder architecture of this embodiment, skip connections directly feed the encoded feature maps output from each resolution level during the downsampling encoding process to the corresponding level in the decoder for upsampling at the same resolution, and then concatenate and fuse them along the channel dimension.

[0040] The skip connection concatenation operation can preserve the independent information channels of the two feature paths without attenuation, allowing subsequent convolutional layers to adaptively learn how to balance the contribution of semantics and details. This allows for the gradual and refined restoration of the segmentation boundary during the upsampling process, ultimately ensuring that the output segmentation mask map achieves the optimal balance between contour accuracy and structural integrity.

[0041] The multi-channel output layer is a convolutional layer located at the final stage in a semantic segmentation network. It is used to transform the number of channels in the final decoded feature map into the total number of target categories, where each channel corresponds to a probability response map with a specific structure.

[0042] Oocyte quality assessment requires the simultaneous quantification of multiple independent morphological structures, such as the zona pellucida, polar body, perivitelline space, cytoplasmic region, and background. A multi-channel design allows the network to simultaneously output pixel-level localization results for all targets in a single forward propagation, improving inference efficiency while ensuring alignment of structural masks in spatial coordinates. This provides a precise spatial reference for subsequent calculations of cross-structural parameters such as thickness ratio and area symmetry.

[0043] Pixel classification refers to the output layer generating a probability vector of length C at each pixel location. After being normalized by Softmax (normalized exponential function), the values ​​of each dimension in the vector represent the confidence that the pixel belongs to the corresponding category. The network selects the category corresponding to the highest confidence as the final semantic label of the pixel, thereby discretizing continuous pixel values ​​into category labels.

[0044] The segmentation mask is a label image of the same size as the original image, which is formed by rearranging the above pixel classification results according to their spatial positions. If multiple label images are visualized, the probability maps of each channel are thresholded and binarized to independently present the region occupied by a certain structure.

[0045] In this embodiment, the multi-channel output layer preferably has 4 channels, such as the first channel outputting a black background mask, the second channel outputting the oocyte region, the third channel outputting the polar body region, and the fourth channel outputting the zona pellucida region. The specific number of channels can be increased or decreased according to the morphological requirements.

[0046] The mask for the black background channel is a completely black background with no effective white contours. The other three channels have white foreground contours and are target channel masks that can be used for contour fitting. The black background mask is used to distinguish between image blank irrelevant pixels and three types of target pixels: oocytes, polar bodies, and zona pellucida. Edge noise is filtered out, and contour fitting and parameter calculation are limited to the effective cell region.

[0047] In the segmentation mask image of the oocyte, polar body, and zona pellucida, white pixels delineate the occupied areas and contour boundaries of the corresponding structures, providing the most basic and accurate pixel-level measurement basis for subsequent geometric parameter calculation based on contour fitting algorithms.

[0048] Contour fitting image algorithms are computational geometry techniques that use discrete boundary pixel sets extracted from segmented mask images to fit these discrete points into smooth, continuous geometric parametric equations through mathematical optimization strategies such as least squares, random sampling consistency, or elliptic Fourier descriptors. For example, the overall contour of an oocyte can be fitted as an ellipse, the inner and outer boundaries of the zona pellucida can be fitted as concentric ellipses, and the polar body region can be fitted as an irregular closed curve.

[0049] Morphological characteristic parameters are numerical indicators derived from fitting equations that can quantitatively characterize the physical structure and geometric properties of oocytes. Specifically, they may include major and minor axis lengths, area, perimeter, roundness or eccentricity, and centroid coordinates obtained from ellipse fitting; average thickness, maximum and minimum thickness values, and thickness variation coefficient obtained from ellipse fitting inside and outside the zona pellucida; area proportion, maximum Feret diameter, mean width of the perivitelline space, and left-right symmetry deviation obtained from fitting the polar body region.

[0050] The segmentation mask image is a discrete pixel grid. The contour fitting image algorithm can improve measurement accuracy and suppress the impact of discretization error and local outliers on the results by introducing global geometric constraints and mathematical smoothing optimization. The objective and continuous standardized values ​​are the basic inputs necessary for the preset oocyte scoring model to carry out accurate regression or classification evaluation, which determines the scientificity and reliability of the final quality evaluation results.

[0051] A pre-defined oocyte scoring model is essentially a decision rule or mathematical model that maps extracted morphological feature parameters to quality assessment results. This pre-defined oocyte scoring model can be either a deterministic scoring rule system based on expert experience, or a regression or classification prediction model trained from a large number of labeled samples using machine learning algorithms.

[0052] In this embodiment, the process of obtaining quality assessment results through morphological feature parameters is a feature-score mapping. Specifically, each morphological parameter is first converted into a sub-score of each indicator according to preset rules or model reasoning. Then, a comprehensive decision-making mechanism such as weighted summation, logistic regression, decision tree, or neural network is used. These sub-scores are aggregated into a total score or quality grade that reflects the overall developmental potential of the oocyte. Finally, based on the comparison between this score and a preset threshold, a conclusive assessment result is output regarding whether the oocyte is suitable for subsequent assisted reproductive procedures.

[0053] The present invention provides an oocyte quality assessment method and system that employs an image segmentation model and downsamples and encodes multiple residual convolutional blocks to extract multi-level semantic features from shallow to deep layers in oocyte images, effectively addressing the problem of small grayscale differences. By extracting multi-scale contextual features in parallel using multiple dilated convolutional kernels with different dilation rates and then splicing and fusing them, a larger receptive field can be obtained without sacrificing spatial resolution. This allows the model to fully perceive the global positional relationships and contextual information between various structures of the oocyte, thereby accurately distinguishing structural regions with similar grayscale but different morphologies.

[0054] Simultaneously, by using skip connections to fuse the encoded feature maps from the downsampling encoding stage with the upsampling decoding results, the loss of spatial information during downsampling is compensated for, resulting in more refined and complete segmentation boundaries. Finally, a multi-channel output layer outputs segmentation mask maps of various morphological features, achieving simultaneous and accurate segmentation of multiple morphological features. The segmentation mask maps are then used to calculate morphological feature parameters through a contour fitting algorithm, which are then input into a scoring model to obtain objective and accurate oocyte quality assessment results. This solves the problem of inaccurate oocyte image structure segmentation in related technologies, which affects the accuracy of oocyte quality assessment.

[0055] As an optional approach, morphological characteristic parameters are input into a preset oocyte scoring model to obtain oocyte quality assessment results, including: determining oocyte assessment indicators based on morphological characteristic parameters; wherein, assessment indicators include oocyte diameter, polar body ellipticity, polar body area, and perivitelline space; obtaining indicator assessment scores according to the scoring criteria corresponding to the assessment indicators; inputting the indicator assessment scores into the preset oocyte scoring model to obtain a comprehensive assessment score; and comparing the comprehensive assessment score with a preset threshold to obtain the quality assessment result.

[0056] In assisted reproductive technology (ART) clinical practice, oocyte diameter reflects oocyte maturity, polar body ellipticity and polar body area together characterize the morphological integrity of the first polar body, and perivitelline space reflects the degree of cytoplasmic contraction and zona pellucida separation during oocyte maturation. These four indicators cover the most representative morphological features of the oocyte's extracytoplasmic structures and can comprehensively reflect the oocyte's developmental potential from multiple dimensions.

[0057] Evaluation metrics refer to specific morphological dimensions used to assess oocyte quality. After processing the image using a contour fitting algorithm, the segmentation module has obtained morphological feature parameters such as oocyte diameter, polar body ellipticity, polar body area, and perivitelline space. Based on these parameters, four evaluation metrics are determined. By transforming the discrete pixel information of the segmentation mask image into these four standardized evaluation metrics, the transformation from subjective morphological observation to objective quantitative measurement is achieved, providing structured and computable input data for the subsequent scoring model.

[0058] Based on years of experience by embryologists in assessing oocyte quality, a pre-defined scoring standard was established for each assessment indicator. Each indicator was then converted into a score using the same metric, ensuring comparability and compatibility among the indicators.

[0059] Different morphological features have varying degrees of influence on oocyte developmental potential; for example, polar body morphology contributes more to the assessment results than perivitelline space. By assigning different weighting coefficients to each indicator, the actual importance of each morphological feature in clinical assessment can be reflected more accurately.

[0060] The pre-built oocyte scoring model is constructed based on embryologists' assessment experience. It employs a weighted summation comprehensive decision-making mechanism to aggregate the scores of four individual indicators and output a comprehensive assessment score. The four independent morphological dimensions are integrated into a quantitative indicator that can comprehensively reflect the oocyte's developmental potential, thereby solving the problem of the one-sidedness of single indicator assessment.

[0061] In assisted reproductive technology (ART) clinical procedures, embryologists need to select the highest quality oocytes from multiple sources for ICSI injection. By setting clear thresholds, the continuous composite score is transformed into a discrete operational decision, providing clear and actionable screening criteria for clinical practice.

[0062] In this embodiment, the preset threshold is set based on the assessment experience of embryologists. If the comprehensive assessment score is greater than or equal to the preset threshold, the oocyte is determined to be of good quality and suitable for subsequent assisted reproductive operations such as IVF and ICSI. If the comprehensive assessment score is lower than the preset threshold, the oocyte is determined to be of unsuitable quality for subsequent assisted reproductive operations.

[0063] This embodiment transforms abstract numerical scores into specific clinical operation instructions. Based on the assessment results, operators can directly select the oocytes with the highest quality assessment scores for subsequent assisted reproductive operations, thereby achieving a complete closed loop from quantitative analysis to clinical decision-making in oocyte quality assessment.

[0064] As an alternative approach, assessment indicators for oocytes are determined based on morphological characteristic parameters, including the following expression for polar body ellipticity: Where e represents the ellipticity of the polar body, a represents the major axis of the elliptical profile of the polar body, and b represents the minor axis of the elliptical profile of the polar body; the expression for the area of ​​the polar body is as follows: Where s represents the polar body area and p represents the pixel distance calibration value; the expression for the periovarian gap is as follows: Where d represents the periovarian space, D represents the diameter of the inner ring of the zona pellucida, and D represents the diameter of the oocyte.

[0065] Under a microscope, the first polar body naturally exhibits an approximately elliptical shape. The boundary obtained by the direct pixel counting method is a jagged, stepped shape, which cannot accurately reflect the true geometry of the polar body. By using ellipse fitting to transform the discrete boundary pixels into a continuous, smooth elliptical equation, the discretization error caused by the pixel grid can be eliminated, and the degree to which the polar body deviates from a perfect circle can be quantified.

[0066] The major axis 'a' is defined as the length of the fitted ellipse along its longest diameter, and the minor axis 'b' is defined as the length of the fitted ellipse along its shortest diameter. When the polar body is perfectly circular, a = b, and e = 1. When the polar body deviates from a perfect circle, the ratio of b to a deviates from 1, and the degree to which the value of e deviates from 1 directly reflects the degree of ellipticization of the polar body shape.

[0067] The calculation π×a×b yields the area of ​​the fitted ellipse in the image pixel coordinate system. This area is then multiplied by p², where p is the pixel distance calibration value in μm / pixel, converting the pixel area to the actual physical area in μm². The pixel distance calibration value p is predetermined by the calibration process of the microscopic imaging system, for example, by photographing a calibration micrometer of known size to establish the correspondence between the number of pixels and the actual number of micrometers.

[0068] The actual physical size represented by a single pixel varies under different microscopic imaging systems and objective magnifications. By converting the area to μm² using p², a unit of measurement with absolute physical significance, the comparability of oocyte morphology measurement results obtained from different batches and under different imaging conditions is ensured.

[0069] D is the diameter of the inner ring of the zona pellucida, and D is the diameter of the oocyte. The difference between the two is ( -D) represents the total radial gap width between the inner wall of the zona pellucida and the outer edge of the oocyte, which is the sum of the gaps on both sides. Dividing by 2 gives the width of the periovarian gap on one side. Multiplying by the pixel distance calibration value p gives the periovarian gap value, converting the pixel distance into the actual physical width in μm.

[0070] The perivitelline space is an important morphological parameter characterizing oocyte maturity; both excessively wide and narrow spaces indicate abnormal oocyte development. The value of the perivitelline space provides a quantitative representation of oocyte maturity for scoring models.

[0071] This embodiment transforms the discrete pixel information of the segmentation mask image into geometric parameters with clear physical meaning. The polar body shape is standardized and quantified through polar body ellipticity; the polar body area converts pixel area into absolute physical area, thus achieving cross-sample comparability; and the perioval gap accurately converts the diameter difference into the actual gap width. These three indicators provide objective, quantifiable, and repeatable numerical inputs for the subsequent scoring model, solving the problems of strong subjectivity, low pixel-level measurement accuracy, and incomparability in manual evaluation.

[0072] As an optional approach, the segmentation mask image includes the oocyte, polar body, and zona pellucida; morphological feature parameters include the oocyte diameter, the major and minor axes of the polar body, and the inner ring diameter of the zona pellucida; based on the segmentation mask image, the morphological feature parameters of the oocyte are obtained through contour fitting image algorithms, including: based on the oocyte segmentation mask image, the oocyte diameter is obtained through circular contour fitting; based on the polar body segmentation mask image, the major and minor axes of the elliptical contour of the polar body are obtained through elliptical contour fitting; based on the zona pellucida segmentation mask image, the inner ring diameter of the zona pellucida is obtained through annular contour fitting.

[0073] Oocytes exhibit an approximately circular geometric shape in microscopic images. The contour fitting module extracts the set of boundary pixels from the segmented mask image of the oocyte and processes these discrete boundary points using a circular contour fitting algorithm. The goal is to find an optimal circle that achieves the best match between the circle and all pixels on the mask boundary in a least-squares sense, thus obtaining the diameter parameter of the fitted circle, i.e., the oocyte diameter. The circular contour fitting can be implemented using methods such as least-squares circle fitting or geometric distance minimization.

[0074] The polar body exhibits an approximately elliptical geometric shape in the microscopic image. The contour fitting module extracts the boundary pixel set from the segmented mask image of the polar body and processes these discrete boundary points using an elliptical contour fitting algorithm. An optimal ellipse is found that achieves the best match between the ellipse and all pixels on the polar body mask boundary in the least-squares sense, thus obtaining the two key geometric parameters of the fitted ellipse: the major axis and the minor axis. The elliptical contour fitting can be implemented using methods such as eigenvalue decomposition based on the covariance matrix and direct least-squares elliptical fitting.

[0075] The transparent band exhibits a ring-shaped geometry in microscopic images. The contour fitting module extracts the pixel sets of the inner and outer boundaries from the segmented mask image of the transparent band and processes them using a ring contour fitting algorithm. Specifically, the inner boundary of the transparent band is fitted as an inner circle, and the outer boundary as an outer circle, and the two are usually considered as concentric circles or approximately concentric circles, thereby obtaining the diameter parameter of the inner ring of the transparent band. The specific implementation of the ring contour fitting can be achieved by performing circular fitting on the inner and outer boundaries separately and then optimizing with concentric constraints, or by using a ring-based global least squares fitting method.

[0076] This embodiment improves measurement accuracy by converting the discrete boundary pixel set of the segmentation mask into a smooth geometric parametric equation, thereby suppressing the interference of jagged edges and local defects of the segmentation mask boundary on the measurement results and ensuring the cross-sample repeatability and measurement accuracy of morphological feature parameters.

[0077] As an alternative approach, the expression for the oocyte scoring model is as follows: ;in, The index score representing the diameter of the oocyte. The index score representing the ellipticity of the polar body. This represents the evaluation score corresponding to the polar body area. This indicates the assessment score of the index corresponding to the periovarian space.

[0078] In this embodiment, the oocyte scoring model adopts a linear weighted summation form, and the weight coefficients of each indicator are set based on the years of experience of embryologists in assessing oocyte quality.

[0079] The polar body ellipticity and polar body area each account for 30% of the weight, totaling 60%, reflecting the importance of polar body morphology in oocyte quality assessment. The polar body is a marker of oocyte completion of meiosis, and ellipticity and area directly reflect whether the meiotic process is normal.

[0080] The oocyte diameter accounts for 25% of the weighting, reflecting the influence of oocyte size on developmental potential. Both excessively large and small diameters may indicate abnormal cytoplasmic maturity.

[0081] The periovarian space accounts for 15% of the weighting and serves as a supplementary reference indicator. An excessively wide or narrow space may reflect abnormal cytoplasmic contraction and zona pellucida separation during oocyte maturation. However, its weight in clinical assessment is relatively lower than that of polar body morphology and oocyte size. The above weighting allocation is based on years of experience by embryologists in assessing oocyte quality. The sum of the four weights is 1, ensuring that the overall assessment score falls within the scoring range and is consistent with the scoring scale of each individual indicator.

[0082] The linear weighted summation model in this embodiment has a simple structure and strong interpretability. Each weight coefficient corresponds to a clear clinical meaning, making it easy for clinical embryologists to understand and accept. The parameters of this model can be directly set and adjusted based on clinical experience without the need for training with a large number of labeled samples. At the same time, the linear weighted model has better generalization stability and clinical generalizability than complex data-driven models.

[0083] As an optional approach, the evaluation score of the evaluation index is obtained according to the scoring criteria corresponding to the evaluation index, including: setting a benchmark range according to the quality assessment requirements of oocytes; determining the evaluation score of the evaluation index as full marks when the evaluation index is within the benchmark range; and subtracting a preset score from the full marks when the evaluation index exceeds the benchmark range based on the deviation between the evaluation index and the benchmark range. The preset score is proportional to the absolute value of the deviation, and the preset score is greater than zero and less than or equal to the full marks.

[0084] The original numerical dimensions of the four morphological characteristic parameters are different: diameter is in units of length, ellipticity is a dimensionless ratio, and area is in units of area, making direct comparison or synthesis impossible. This embodiment uses a unified 0-10 scoring standard to map morphological characteristics of different dimensions to the same scoring scale, making the indicators comparable and synergistic. Simultaneously, a linear penalty mechanism in the deduction rules, where larger deviations result in higher deductions, achieves quantitative punishment for abnormal morphological characteristics, ensuring that the scoring results accurately reflect the negative impact of morphological deviations on oocyte quality.

[0085] In this embodiment, the scoring criteria corresponding to the oocyte evaluation indicators are pre-set scoring rules, which are constructed based on the years of experience of embryologists in evaluating oocyte quality.

[0086] The pre-defined scoring criteria are shown in Table 1:

[0087] The scoring criteria first establish a benchmark range for each evaluation indicator. When the measured value of an evaluation indicator falls within this benchmark range, the indicator receives a maximum score of 10 points. When the measured value of an evaluation indicator exceeds the benchmark range, a pre-set score is deducted from the maximum score based on the degree of deviation between the measured value and the benchmark range. This pre-set score is proportional to the absolute value of the deviation, and the minimum score after deduction is 0 points.

[0088] Specifically, the baseline range for oocyte diameter is 113 μm to 114 μm, with 2 points deducted for every 1 μm deviation from this range; the baseline range for polar body ellipticity is 0 to 2.5, with 2 points deducted for every 0.1 deviation from this range; the baseline range for polar body area is 150 μm² to 300 μm², with 2 points deducted for every 1 μm² deviation from this range; and the baseline range for periovarian space is 4 μm to 6 μm, with 2 points deducted for every 1 μm deviation from this range.

[0089] Each indicator uses the range of values ​​considered normal in clinical practice as the full score range. The closer the measured value is to the center of the range, the higher the score; the further the measured value deviates from the range, the lower the score. In practical applications, the deduction gradient can be adjusted according to clinical needs to change the sensitivity of the score.

[0090] This embodiment uses a unified scoring framework that combines a benchmark range with linear deduction to accurately map four morphological feature parameters with different dimensions to the same scoring scale. Simultaneously, a linear penalty mechanism quantifies the degree of morphological deviation. Through this standardized transformation, the evaluation scores of each indicator become directly additive, providing a unified, comparable, and continuous input for subsequent weighted summation models, thereby ensuring the quantification and automation of the entire oocyte quality assessment system.

[0091] As an optional approach, before inputting the oocyte image into the image segmentation model, the method further includes: acquiring multiple oocyte image samples, wherein the image samples include the oocyte, polar body, and zona pellucida; performing geometric transformation and image enhancement on the oocyte image samples to obtain enhanced image samples; wherein the geometric transformation and image enhancement include at least one of rotation, horizontal flipping, vertical flipping, and contrast adjustment; manually labeling the enhanced image samples to generate corresponding labels; dividing the enhanced image samples and corresponding labels into a training set, a validation set, and a test set according to a preset ratio; and training the image segmentation model based on the training set, validation set, and test set to obtain a trained image segmentation model.

[0092] Image samples refer to microscopic images of primordial oocytes used to train image segmentation models, such as... Figure 3 As shown, each image must contain complete morphological structures such as the oocyte, polar body, and zona pellucida.

[0093] This embodiment uses a sample size of 200 to balance the feasibility of obtaining clinical data with the basic needs of model training. Subsequent data augmentation will expand the sample size to meet the requirements of deep learning training. First, 200 oocytes are prepared, and one image is acquired for each oocyte under an inverted microscope. All images must clearly show the morphological features of the oocyte, polar body, and zona pellucida. These acquired images serve as the raw foundation data for subsequent data augmentation and model training.

[0094] This embodiment performs data enhancement processing on 200 collected oocyte image samples, specifically employing methods such as rotation, horizontal flipping, vertical flipping, and contrast adjustment to expand the number of images to 2000. Rotation refers to rotating the image around its center by a certain angle, such as ±90° or ±180°, to simulate the oocyte morphology under different shooting angles. Horizontal and vertical flipping refer to mirroring the image along its vertical and horizontal axes, respectively, to increase the orientational diversity of the samples. Contrast adjustment refers to adjusting the pixel grayscale distribution range of the image, such as increasing or decreasing contrast, to simulate the lighting differences under different microscopic imaging conditions.

[0095] The acquisition conditions of the original 200 oocyte image samples, such as light intensity, oocyte orientation, and focal plane position, have limitations. By using enhancement operations such as rotation, flipping, and contrast adjustment, richer imaging condition variations can be simulated, allowing the model to be exposed to more diverse sample distributions during the learning process, thereby improving the model's adaptability to different imaging conditions and oocytes of different morphologies.

[0096] Image segmentation models, as deep learning models, are essentially trained through supervised learning. These models require a large amount of "image-label" pairing data as training samples. The labels provide the true category information for each pixel and are the sole basis for the model to calculate the loss function and perform backpropagation updates.

[0097] In oocyte segmentation, due to the small grayscale differences and blurred boundaries between oocytes, zona pellucida, and polar bodies, manual annotation by experts with embryological expertise can ensure the accuracy and reliability of the labels, thereby providing high-quality supervision signals for model training.

[0098] In this embodiment, the size of the 2000 enhanced image samples is further adjusted to 256 pixels × 256 pixels to meet the input size requirements of the image segmentation model. Then, the Labelme annotation software is used to manually annotate the 2000 resized images. Annotators need to delineate the boundaries of four categories—black background, oocyte, polar body, and zona pellucida—pixel by pixel on each image, generating corresponding pixel-level labels. After annotation, the annotation file format generated by Labelme is converted using code. The label image is in .png format, which is a single-channel category index image. Each pixel value represents the category number to which the position belongs, so as to facilitate data loading and processing during subsequent image segmentation model training.

[0099] The training set is responsible for teaching the image segmentation model the mapping relationship from images to segmentation masks. The validation set monitors the image segmentation model during training to check for overfitting, i.e., a continuous decrease in the loss on the training set while the loss on the validation set increases, and adjusts hyperparameters or stops training early accordingly. The test set provides a one-time objective evaluation after the image segmentation model is finalized to verify its true generalization ability on completely unseen data. Neither the validation nor the test set participates in gradient updates during training, and the test set is used only once in the final evaluation, thus ensuring the objectivity and fairness of the image segmentation model's performance evaluation.

[0100] In this embodiment, 2000 enhanced image samples and their corresponding .png format labels are divided into training, validation, and test sets according to a 6:2:2 ratio. Specifically, the training set contains 1200 images and their labels, used for training and optimizing the image segmentation model parameters; the validation set contains 400 images and their labels, used for hyperparameter adjustment and overfitting monitoring during training; and the test set contains 400 images and their labels, used for the final performance evaluation after the image segmentation model training is completed.

[0101] During training, the image segmentation model uses images from the training set as input and their corresponding labels as supervision targets. It calculates the difference between the predicted segmentation mask and the true label through forward propagation, and then propagates the loss gradient back to each layer of the network through backpropagation to update the network parameters. The validation set is used at the end of each training cycle to evaluate the performance metrics of the current image segmentation model, such as accuracy, sensitivity, specificity, and Dice. The model is monitored for overfitting, and hyperparameters such as the learning rate are adjusted accordingly, or training is stopped early.

[0102] After training, the final segmentation performance of the image segmentation model is objectively evaluated once using a test set that was never used in the training, and the test results of the image segmentation model performance are obtained.

[0103] The test results of the image segmentation model performance are shown in Table 2:

[0104] The test results in the table show an accuracy of 94%, sensitivity of 90%, specificity of 95%, and Dice coefficient of 92%, representing the evaluation results of a test set of 400 images and their corresponding labels. These evaluation results demonstrate that the image segmentation model trained in this embodiment exhibits significant inference performance and can be used for real-time segmentation inference of oocyte images.

[0105] This embodiment addresses the scarcity of labeled samples in the field of medical imaging at low cost through data augmentation, ensures the high quality and accuracy of training labels through expert manual annotation, and provides a standardized data foundation for model training and performance evaluation through training, validation, and test sets. Finally, a high-precision image segmentation model is obtained through sufficient training.

[0106] The oocyte quality assessment method of this invention automatically acquires oocyte morphological characteristic parameters and quality assessment scores from oocyte images, free from the limitations of embryologists' subjective judgment. Furthermore, this method provides a comprehensive evaluation of oocyte morphological characteristics, reducing the bias inherent in evaluations based on a single morphological feature. Therefore, this oocyte quality assessment method enables objective, accurate, and automated assessment of oocyte quality, thereby improving the success rate of subsequent assisted reproductive procedures.

[0107] According to another aspect of the present invention, an oocyte quality assessment system is also provided for performing the above-described oocyte quality assessment method, such as... Figure 3 As shown, it includes: an acquisition module, a segmentation module, and an evaluation module; the acquisition module is used to acquire oocyte images and input the oocyte images into the segmentation module; the segmentation module is connected to the acquisition module and is used to input the oocyte images into the image segmentation model, which performs downsampling and encoding processing through multiple residual convolutional blocks to obtain encoded feature maps at different scales, and generates the final encoded feature map in the deepest layer; multi-scale contextual features of the final encoded feature map are extracted in parallel through various dilated convolutional kernels with different dilation rates, and the multi-scale contextual features are concatenated and fused to obtain a fused feature map; the fused feature map is then processed... The upsampling decoding result from the upsampling decoding process is concatenated and fused with the corresponding level of the encoded feature map in the downsampling encoding process through skip connections to obtain the final decoded feature map. A multi-channel output layer is used to classify the final decoded feature map by pixels and output a segmentation mask map. This multi-channel output layer outputs segmentation masks representing various morphological features of the oocyte. An evaluation module connected to the segmentation module is used to obtain the morphological feature parameters of the oocyte based on the segmentation mask map using a contour fitting image algorithm. These morphological feature parameters are then input into a preset oocyte scoring model to obtain the quality assessment result of the oocyte.

[0108] The acquisition module refers to the hardware or a combination of hardware and software functional units used to acquire images of oocytes. This module is responsible for acquiring raw microscopic images containing morphological parts such as the oocyte, polar body, and zona pellucida, and transmitting the acquired images to the segmentation module for subsequent processing.

[0109] The acquisition module typically includes image acquisition hardware such as an inverted microscope, digital camera, or industrial camera, along with its corresponding driver and control software. The operator places the oocyte sample to be evaluated on the stage of the inverted microscope and, by adjusting the microscope's objective magnification and focal length, acquires a clear microscopic image of the oocyte. This image must clearly show the morphological features of the oocyte, polar body, and zona pellucida. The acquired image is transmitted to the segmentation module via a data interface as input data for the segmentation module.

[0110] The segmentation module is a functional unit used for real-time semantic segmentation of oocyte images, and it contains a pre-trained image segmentation model. This module takes the oocyte image output from the acquisition module as input, infers from the image segmentation model, outputs a segmentation mask image, and further calculates the morphological feature parameters of the oocyte using a contour fitting image algorithm.

[0111] like Figure 2 As shown, the image segmentation model in this embodiment specifically adopts the improved UNet segmentation network. The symmetrical U-shaped architecture presents the complete information flow path of the segmentation model of the present invention from image input to mask output. The positions and connections of residual structure, skip connections, dilated spatial convolution structure, and multi-channel output layer in the network are marked.

[0112] Starting from the input layer on the left, the original oocyte image first enters the downsampling encoding stage. Through the stacked residual structure, i.e. the residual convolutional block, it is downsampled multiple times in sequence. After each downsampling, the number of channels of the feature map increases step by step.

[0113] like Figure 2 The bottom layers, with 64, 128, 256, 512, and 1024 channels, correspond to the number of channels at each level, with the spatial resolution halved at each level. This allows for the extraction of encoded feature maps at different scales, from shallow details to deep semantics. At the deepest layer, with 1024 channels, after generating the final encoded feature map, a dilated spatial convolution structure (i.e., dilated convolution kernels with multiple dilation rates) is introduced to extract and fuse multi-scale contextual features in parallel, enhancing the network's ability to perceive global information.

[0114] The next step is the upsampling decoding stage, where the spatial resolution of the feature map is gradually restored through upsampling decoding operations. Skip connections are then used to stitch and fuse the feature map of the corresponding level in the encoding stage with the upsampling decoding result in the channel dimension to compensate for the edge information lost during the upsampling process.

[0115] Finally, after a 1×1 convolutional layer, the feature map is mapped to a multi-channel output. In this embodiment, the number of channels is expanded to four, corresponding to the black background, oocyte, polar body, and zona pellucida, respectively. After pixel classification, the output shows three types of morphological structure segmentation mask images as shown on the right: oocyte, polar body, and zona pellucida. The black background mask is completely black with no effective white contours, and therefore contains no contours that can be used for subsequent calculations. Figure 2 Not shown in the image.

[0116] The assessment module is a functional unit used for comprehensive evaluation and decision-making regarding oocyte quality. This module takes the morphological feature parameters output by the segmentation module as input, calculates a comprehensive assessment score by calling a preset oocyte scoring model, compares the comprehensive assessment score with a preset threshold, and outputs the oocyte quality assessment result, indicating whether it is suitable for subsequent assisted reproductive procedures such as IVF and ICSI.

[0117] The assessment module, serving as the decision-making output terminal of the entire system, is responsible for transforming the purely geometric parameters output by the segmentation module into clinically significant conclusions. The assessment module's input is connected to the segmentation module, receiving morphological feature parameters as input data. The assessment module's output is directed to clinical operators, directly providing the final conclusion of oocyte quality assessment, thus forming a complete assessment loop. By embedding the scoring logic and decision rules within the assessment module, it ensures that every quality assessment strictly adheres to the same scoring standards and threshold judgment rules, eliminating subjective differences inherent in manual assessment.

[0118] To further test the accuracy of the oocyte quality assessment method of this invention, 300 oocytes were divided into three groups: a manual observation group, an image thresholding group, and the oocyte quality assessment group of this invention. During ICSI (Intracytoplasmic Sperm Injection), the oocyte quality was inferred using these three methods, and a comparison of fertilization rate and blastocyst formation rate under different oocyte quality assessment methods was obtained.

[0119] Table 3 shows a comparison of fertilization rates and blastocyst formation rates for ICSI procedures using different oocyte quality assessment methods.

[0120] The results showed that the oocyte quality assessment method evaluated in this invention had the highest fertilization rate and blastocyst formation rate, reaching 97.6% and 79.8%, respectively; while the image thresholding method had 91.4% and 78.5%, and the manual observation method had 85.2% and 71.6%, with lower fertilization rates and blastocyst formation rates.

[0121] The above data fully demonstrates that the technical solution of this invention, which uses a deep learning segmentation model to accurately extract morphological feature parameters and combines them with a preset scoring model for comprehensive quantitative evaluation, can more accurately screen oocytes with high developmental potential than traditional subjective judgment and image threshold segmentation methods, thereby effectively improving the success rate of ICSI and the quality of embryo development. This experimental result also provides strong empirical support for the objective, accurate, and automatic assessment of oocyte quality and the improvement of the success rate of assisted reproductive technologies.

[0122] According to another aspect of the invention, a non-transitory machine-readable medium storing computer instructions for causing a computer to perform any of the methods described above is also provided.

[0123] An embodiment of the present invention also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, which, when executed by the at least one processor, causes the electronic device to perform the method of the embodiment of the present invention.

[0124] refer to Figure 5 The present invention will now describe a structural block diagram of an electronic device that can serve as an embodiment of the present invention, serving as an example of a hardware device applicable to various aspects of the present invention. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0125] like Figure 5 As shown, the electronic device includes a computing unit 501, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 502 or a computer program loaded from a storage unit 508 into a random access memory (RAM) 503. The RAM 503 may also store various programs and data required for the operation of the electronic device. The computing unit 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0126] Multiple components in the electronic device are connected to I / O interface 505, including: input unit 506, output unit 507, storage unit 508, and communication unit 509. Input unit 506 can be any type of device capable of inputting information into the electronic device. Input unit 506 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of the electronic device. Output unit 507 can be any type of device capable of presenting information and may include, but is not limited to, a display, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 508 may include, but is not limited to, disks and optical discs. Communication unit 509 allows the electronic device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, and / or wireless communication transceivers, such as Bluetooth devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0127] The computing unit 501 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, CPUs, graphics processing units (GPUs), various special-purpose artificial intelligence (AI) computing units, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. The computing unit 501 performs the various methods and processes described above. For example, in some embodiments, the method embodiments of the present invention can be implemented as computer programs tangibly contained in a machine-readable medium, such as storage unit 508. In some embodiments, part or all of the computer program can be loaded and / or installed on an electronic device via ROM 502 and / or communication unit 509. In some embodiments, the computing unit 501 can be configured to perform the methods described above by any other suitable means (e.g., by means of firmware).

[0128] Computer programs for implementing the methods of embodiments of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0129] In the context of embodiments of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable signal medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, or infrared systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0130] It should be noted that the term "comprising" and its variations used in the embodiments of this invention are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The modifications of "one" and "a plurality" mentioned in the embodiments of this invention are illustrative and not restrictive, and those skilled in the art should understand that unless explicitly indicated otherwise in the context, they should be understood as "one or more".

[0131] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of this invention are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0132] The steps described in the method embodiments provided by the present invention can be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of protection of the present invention is not limited in this respect.

[0133] The term "embodiment" in this specification refers to a specific feature, structure, or characteristic described in connection with an embodiment that may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily imply the same embodiment, nor does it imply independence or alternativeity from other embodiments. The various embodiments in this specification are described in a related manner, with reference to each other for similar or identical parts. In particular, for apparatus, device, and system embodiments, since they are substantially similar to method embodiments, the description is relatively simple, and relevant details are referred to in the description of the method embodiments.

[0134] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of protection. It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A method for assessing oocyte quality, characterized in that, include: The oocyte image is input into the image segmentation model, and then processed by multiple residual convolutional blocks for downsampling and encoding to obtain encoded feature maps at different scales. Finally, the encoded feature map is generated in the deepest layer. Multi-scale contextual features of the final encoded feature map are extracted in parallel using dilated convolution kernels with different dilation rates, and then the multi-scale contextual features are concatenated and fused to obtain a fused feature map. The upsampled decoding result of the fused feature map is combined with the corresponding level of the encoded feature map in the downsampled encoding by skip connections to obtain the final decoded feature map. The final decoded feature map is classified by pixels through a multi-channel output layer, and a segmentation mask map is output; wherein, the multi-channel output layer outputs segmentation mask maps of various morphological features of the oocyte. Based on the segmentation mask image, the morphological feature parameters of the oocyte are obtained through a contour fitting image algorithm. The morphological feature parameters are then input into a preset oocyte scoring model to obtain the quality assessment result of the oocyte.

2. The method for assessing oocyte quality according to claim 1, characterized in that, The morphological characteristic parameters are input into a preset oocyte scoring model to obtain the quality assessment results of the oocytes, including: Based on the morphological characteristic parameters, the evaluation indicators of the oocyte are determined; wherein, the evaluation indicators include oocyte diameter, polar body ellipticity, polar body area, and perivitelline space; Based on the scoring criteria corresponding to the evaluation indicators, the evaluation scores corresponding to the evaluation indicators are obtained. The scores of the aforementioned indicators are input into a preset oocyte scoring model to obtain a comprehensive evaluation score; The overall evaluation score is compared with a preset threshold to obtain the quality evaluation result.

3. The oocyte quality assessment method according to claim 2, characterized in that, Based on the morphological characteristic parameters, the evaluation indicators of the oocyte are determined, including: The expression for the polar body ellipticity is as follows: ; Where e represents the ellipticity of the polar body, a represents the major axis of the elliptical profile of the polar body, and b represents the minor axis of the elliptical profile of the polar body. The expression for the area of ​​the polar body is as follows: ; Where s represents the polar body area and p represents the pixel distance calibration value; The expression for the periovarian space is as follows: ; Where d represents the periovarian space, D represents the diameter of the inner ring of the zona pellucida, and D represents the diameter of the oocyte.

4. The method for assessing oocyte quality according to claim 1, characterized in that, The segmentation mask image includes a black background, an oocyte, a polar body, and a zona pellucida; the morphological feature parameters include the oocyte diameter, the long and short axes of the polar body, and the inner ring diameter of the zona pellucida; based on the segmentation mask image, the morphological feature parameters of the oocyte are obtained through a contour fitting image algorithm, including: Based on the segmentation mask image of the oocyte, the diameter of the oocyte is obtained by fitting a circular contour. Based on the segmentation mask image of the polar body, the major axis and minor axis of the elliptical contour of the polar body are obtained by fitting the elliptical contour. Based on the segmented mask image of the transparent band, the inner ring diameter of the transparent band is obtained by fitting the annular contour.

5. The method for assessing oocyte quality according to claim 2, characterized in that, The expression for the oocyte scoring model is as follows: ； in, The index score representing the diameter of the oocyte. The index score representing the ellipticity of the polar body. This represents the evaluation score corresponding to the polar body area. This indicates the assessment score of the index corresponding to the periovarian space.

6. The method for assessing oocyte quality according to claim 2, characterized in that, Based on the scoring criteria corresponding to the evaluation indicators, the evaluation scores for the evaluation indicators are obtained, including: Based on the oocyte quality assessment requirements, a benchmark range is set; If the evaluation indicator is within the benchmark range, the evaluation score of the corresponding indicator is determined to be full marks; If the evaluation index exceeds the benchmark range, a preset score is subtracted from the full score based on the deviation between the evaluation index and the benchmark range; wherein the preset score is proportional to the absolute value of the deviation, and the preset score is greater than zero and less than or equal to the full score.

7. The method for assessing oocyte quality according to claim 1, characterized in that, Before inputting the oocyte image into the image segmentation model, the following steps are also included: Multiple oocyte image samples were acquired, wherein the image samples included the oocyte, polar body, and zona pellucida. The oocyte image sample is subjected to geometric transformation and image enhancement to obtain an enhanced image sample; wherein the geometric transformation and image enhancement include at least one of rotation, horizontal flipping, vertical flipping and contrast adjustment. The enhanced image samples are manually labeled to generate corresponding tags; The enhanced image samples and their corresponding labels are divided into a training set, a validation set, and a test set according to a preset ratio; The image segmentation model is trained using the training set, validation set, and test set to obtain a trained image segmentation model.

8. A system for assessing oocyte quality, characterized in that, The method for performing oocyte quality assessment as described in claim 1 includes: a collection module, a segmentation module, and an assessment module; The acquisition module is used to acquire oocyte images and input the oocyte images into the segmentation module; The segmentation module is connected to the acquisition module and is used to input the oocyte image into the image segmentation model. The image is then processed sequentially through multiple residual convolutional blocks for downsampling encoding to obtain encoded feature maps at different scales, and a final encoded feature map is generated at the deepest layer. Multi-scale contextual features of the final encoded feature map are extracted in parallel using various dilated convolutional kernels with different dilation rates, and these multi-scale contextual features are concatenated and fused to obtain a fused feature map. The upsampling decoding result of the fused feature map is then concatenated and fused with the corresponding level of the encoded feature map in the downsampling encoding through skip connections to obtain the final decoded feature map. Finally, a multi-channel output layer performs pixel classification on the final decoded feature map and outputs a segmentation mask map. The multi-channel output layer outputs segmentation mask maps corresponding to various morphological features of the oocyte. The evaluation module is connected to the segmentation module and is used to obtain the morphological feature parameters of the oocyte based on the segmentation mask image and through a contour fitting image algorithm. The morphological feature parameters are then input into a preset oocyte scoring model to obtain the quality evaluation result of the oocyte.

9. An electronic device, comprising: A processor and a memory storing a program, characterized in that the program includes instructions that, when executed by the processor, cause the processor to perform the method according to any one of claims 1 to 7.

10. A non-transitory machine-readable medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 7.

Images