Image detection and evaluation method for thawing morphology of fertilized egg embryos

Abstract

This invention discloses an image detection and evaluation method for the morphology of thawed fertilized egg embryos, comprising the following steps: acquiring a time-series image sequence of thawed fertilized egg embryos; performing image standardization processing on the time-series image sequence to generate a standardized embryo image set; extracting multi-dimensional embryo quality parameters from the standardized embryo image set; wherein, the multi-dimensional embryo quality parameters include a set of static morphological parameters, a set of dynamic change parameters, and a set of texture feature parameters; generating the blastocyst formation rate based on the multi-dimensional embryo quality parameters; and outputting a three-level embryo quality assessment result based on the comparison relationship between the blastocyst formation rate and a preset threshold range; wherein, the three-level embryo quality assessment result includes a good embryo grade, a qualified embryo grade, and an abnormal embryo grade. This invention has the following advantages and effects: it achieves fully automated, multi-dimensional quantitative evaluation of the thawed embryo recovery process, thereby improving the scientific nature of embryo selection and the clinical pregnancy success rate.

Description

Technical Field

[0001] This invention relates to the field of image analysis technology, and in particular to an image detection and evaluation method for the morphology of thawed fertilized egg embryos. Background Technology

[0002] In the field of assisted reproductive technology, especially in in-vitro fertilization (IVF), embryo quality assessment is a crucial step in determining implantation success rates and pregnancy outcomes. High-quality embryos are more likely to develop into blastocysts and ultimately achieve a successful pregnancy; therefore, accurate embryo quality assessment to select the best embryos for transfer is essential. Traditionally, embryo quality assessment relies on embryologists manually observing embryo morphological characteristics (such as cell number, symmetry, and degree of fragmentation) under a microscope. However, this method has significant limitations, especially when assessing thawed embryos.

[0003] With the widespread use of frozen-thawed embryos in in-vitro fertilization cycles, treatment offers time flexibility and reduces the risk of ovarian hyperstimulation syndrome. However, the quality assessment of thawed embryos faces unique challenges. The freezing and thawing process can affect the structural integrity and developmental potential of the embryo, leading to morphological changes, delayed recovery, or abnormal cell division patterns. These dynamic changes are difficult to fully capture in traditional static morphological assessments. Furthermore, manual assessment methods are not only highly subjective and time-consuming, but also lack consistency due to differences in embryologists' experience and judgment criteria. In particular, the morphological and recovery characteristics of thawed embryos may differ significantly from those of fresh embryos, making it difficult for traditional methods to accurately reflect their true developmental potential.

[0004] Therefore, there is an urgent need for an objective, efficient, and standardized method to assess the quality of thawed embryos, overcoming the shortcomings of traditional methods in terms of accuracy and efficiency. This method should be able to comprehensively analyze the morphological characteristics, dynamic behavior, and texture properties of thawed embryos, thereby providing more reliable quality assessment results, improving embryo selection efficiency, and optimizing pregnancy success rates. This invention addresses this core issue by proposing an image detection and assessment method for the morphology of thawed fertilized egg embryos, achieving accurate assessment of embryo quality through automated and multi-dimensional analysis. Summary of the Invention

[0005] The purpose of this invention is to provide an image detection and evaluation method for the morphology of thawed fertilized eggs and embryos, in order to solve the problems mentioned in the background art.

[0006] The above-mentioned technical objective of the present invention is achieved through the following technical solution:

[0007] A method for image detection and evaluation of the morphology of thawed fertilized egg embryos includes the following steps:

[0008] S100. Obtain a time-series image sequence of fertilized egg embryos after thawing; wherein, the time-series image sequence includes multiple frames of microscopic images of fertilized egg embryos acquired at fixed time intervals within a predetermined time period in the thawing and resuscitation culture medium;

[0009] S200. Perform image standardization processing on the time-series image sequence to generate a standardized embryo image set; wherein, the image standardization processing includes background noise filtering, illumination uniformity correction, and image resolution unification.

[0010] S300. Extract multi-dimensional embryo quality parameters from the standardized embryo image set; wherein, the multi-dimensional embryo quality parameters include a set of static morphological parameters, a set of dynamic change parameters, and a set of texture feature parameters;

[0011] S400. Generate blastocyst formation rate based on the multi-dimensional embryo quality parameters;

[0012] S500. Based on the comparison between the blastocyst formation rate and the preset threshold range, output the three-level embryo quality assessment results; wherein, the three-level embryo quality assessment results include excellent embryo grade, qualified embryo grade and abnormal embryo grade.

[0013] By employing the aforementioned technical solutions, and acquiring time-series image sequences of thawed fertilized embryos, the dynamic changes of embryos during the thawing and recovery process can be captured, providing a rich data base for quality assessment. Image standardization of the time-series image sequences helps improve image quality, reduces errors caused by different shooting conditions, and provides consistent data for subsequent analysis. Extracting multi-dimensional embryo quality parameters from the standardized embryo image set enables a comprehensive assessment of embryo quality characteristics, reflecting the embryo's health status from multiple perspectives. Specifically, extracting a set of morphological static parameters precisely quantifies the embryo's geometric structure and compositional features; a set of dynamic change parameters captures the temporal patterns of embryonic contraction and recovery processes; and a set of texture feature parameters reveals the subtle structural features within the cytoplasm, thus achieving comprehensive coverage of multi-dimensional embryo quality characteristics. Generating blastocyst formation rate based on these multi-dimensional embryo quality parameters can predict the likelihood of embryos developing into blastocysts, providing an important reference for clinical decision-making. Based on the comparison between the blastocyst formation rate and a preset threshold range, a three-tiered embryo quality assessment result is output, facilitating embryo quality grading and enabling clinicians to select the highest quality embryos for transplantation, thereby improving the transplantation success rate.

[0014] A further provision is that the image normalization process for the time-series image sequence in S200 includes:

[0015] Perform background noise filtering operation: For each frame of microscopic image in the time-series image sequence, apply a preset Gaussian convolution kernel for spatial domain filtering, and dynamically calculate the noise intensity threshold based on the difference in gray values ​​of adjacent pixels. Pixels below the noise intensity threshold are identified as background noise and filled with morphological closing operation to generate a denoised intermediate image set.

[0016] Perform illumination uniformity correction operation: use the multi-scale Retinex algorithm to calculate the luminance component of each frame image in the denoised intermediate image set, and calculate the illumination uniformity index on the luminance component. If the illumination uniformity index is lower than the set threshold, start gamma correction to adjust the global contrast and generate an illumination-equalized image set.

[0017] Perform image resolution normalization operation: Set a baseline resolution, calculate the scaling factor for the illumination equalization image set, and then use a bicubic interpolation algorithm to scale the image set according to the scaling factor to generate a resolution normalized image set.

[0018] By adopting the above technical solutions, Gaussian convolution kernel filtering combined with dynamic noise intensity threshold calculation is applied to effectively distinguish between real embryonic structures and background interference, thereby improving denoising accuracy. Morphological closing operations are used to fill in and eliminate residual noise holes, maintaining the morphological integrity of the embryonic region. The multi-scale Retinex algorithm is used to adaptively separate the illumination component and the reflection component, accurately correcting the phenomenon of uneven illumination. The introduction of illumination uniformity can specifically optimize the global contrast distribution, avoiding over-correction or under-correction problems. A baseline resolution is set and a scaling factor is calculated. Combined with a bicubic interpolation algorithm, resolution is unified to ensure the scale consistency of subsequent feature extraction.

[0019] A further setting is that the extraction process of the morphological static parameter set in S300 includes:

[0020] Based on each frame of microscopic image in the standardized embryo image set, an embryo region segmentation operation is performed; wherein, the embryo region segmentation operation employs a threshold segmentation algorithm combined with edge gradient detection to generate a binary embryo mask; then, the geometric features of the region covered by the binary embryo mask are calculated; wherein, the geometric features include:

[0021] Connected component labeling is performed on the binarized embryo mask, the total number of pixels in the largest connected component is counted and multiplied by the single pixel physical size conversion factor to generate the embryo projection area;

[0022] Based on the minimum bounding rectangle of the largest connected region, the ratio of the major axis length to the minor axis length is calculated as the major-minor axis ratio; at the same time, the ratio of the actual perimeter of the largest connected region to the equivalent circle circumference is calculated as the shape irregularity.

[0023] A ring-shaped detection area is constructed around the largest connected region. The number of gray value abrupt change points in this area and the proportion of the total number of pixels in the ring-shaped area are calculated to obtain the proportion of abnormal points in the transparent band.

[0024] The watershed algorithm is applied to the internal region of the binarized embryo mask to segment blastomeres, and the number of closed regions with independent centroids is counted to obtain the effective number of blastomeres.

[0025] The embryo's projected area, aspect ratio, morphological irregularity, proportion of abnormal points in the zona pellucida, and number of effective blastomeres are combined into the set of morphological static parameters.

[0026] By employing the above technical solutions, a high-precision binary embryo mask is generated through threshold segmentation and edge gradient detection to accurately separate the embryo from the background region; connected component labeling focuses on the largest connected component, eliminating fragment interference and ensuring the effectiveness of the analysis object; the major-minor axis ratio quantifies the degree to which the embryo's morphology deviates from a circle, reflecting structural symmetry; morphological irregularity is indicated by comparing the actual circumference with the equivalent circle circumference, showing the complexity of the contour and potential structural defects; a ring detection area is constructed to statistically analyze the proportion of abnormal points in the zona pellucida, sensitively capturing key abnormalities such as zona pellucida damage or deformation; the watershed algorithm is applied to segment blastomeres and count the effective number, accurately assessing cell division status and embryonic composition integrity; and a multi-parameter integrated morphological static set effectively covers the core geometric and structural characteristics affecting embryonic development.

[0027] A further setting is that the extraction process of the dynamically changing parameter set in S300 includes:

[0028] A temporal alignment operation is performed on the standardized embryo image set; wherein, the temporal alignment operation employs embryo centroid tracking for cross-frame position correction;

[0029] Calculate dynamic change features on the temporally aligned image sequence:

[0030] Within a predetermined time period, a fast Fourier transform is applied to the embryo projection area of ​​consecutive frames to extract the main peak frequency value with the largest amplitude in the spectrum, and this frequency value is limited to within a preset effective frequency band to obtain the contraction motion frequency.

[0031] Starting from the time sequence start frame, continuously monitor the contour area change rate. When the absolute value of the change rate is lower than the preset stable threshold for multiple consecutive windows, record the corresponding timestamp as the recovery delay time.

[0032] Regular grid points are selected within the embryonic cytoplasm region as sampling points. For a continuous time frame sequence, the root mean square error of the gray value difference between corresponding sampling points between each pair of adjacent frames is calculated. The time series formed by the calculated root mean square errors between all adjacent frames is linearly fitted to obtain its slope.

[0033] The contraction frequency, recovery delay time, and change slope are combined into the set of dynamic change parameters.

[0034] By adopting the above technical solutions, the embryo centroid tracking method is used to achieve cross-frame position correction, eliminating the interference of embryo displacement on dynamic analysis; a fast Fourier transform is performed on the embryo projection area sequence to extract the main peak frequency value and limit it to the effective frequency band, accurately quantifying the rhythmic characteristics of contraction movement; the contour area change rate is continuously monitored and a stable threshold is set to objectively determine the delay time for the embryo to recover from the thawed state to the steady state; a regular grid dot matrix is ​​laid out in the cytoplasm region, the root mean square error sequence of gray-level differences between adjacent frames is calculated, and the overall trend direction of cytoplasmic dynamic changes is quantified by linear fitting slope.

[0035] A further setting is that the extraction process of the texture feature parameter set in S300 includes:

[0036] Local region of interest extraction is performed on each frame of the microscopic image of the standardized embryo image set; wherein, the local region of interest is the cytoplasmic region inside the embryo excluding the blastomeres;

[0037] Calculate texture features on the local region of interest; wherein the texture features include:

[0038] The texture spectrum of a local region of interest in the embryonic cytoplasm is calculated by applying a rotation-invariant local binary mode operator; the cumulative probability of high-frequency modes in the texture spectrum is statistically analyzed, and the cumulative probability value is used as a quantitative index to describe the graininess intensity of the texture in the local region of interest.

[0039] The region of interest in the embryonic cytoplasm is divided into a predetermined number of rectangular sub-blocks. The average gray value of all pixels within each rectangular sub-block is calculated to generate a set of gray value averages for each sub-block. Based on the set of gray value averages for each sub-block, the standard deviation and arithmetic mean of the gray value averages of all sub-blocks are calculated to obtain the coefficient of variation. The coefficient of variation serves as a quantitative indicator of the uniformity within the cytoplasm. The coefficient of variation is negatively correlated with uniformity.

[0040] Two-dimensional discrete wavelet multi-scale decomposition is performed on the local region of interest; the ratio of low-frequency approximate subband energy to high-frequency detail subband energy is extracted as frequency domain feature, and the Pearson correlation coefficient between horizontal and vertical detail subbands is calculated as texture direction feature.

[0041] By employing the above technical solutions, interference from blastomere boundaries is eliminated, allowing for precise localization of the region of interest in the cytoplasm and focusing on the effective range of texture analysis. A rotation-invariant local binary mode operator is applied to calculate the texture spectrum, and the cumulative probability of high-frequency modes is combined to characterize the intensity of granularity, sensitively capturing the molecular aggregation or particulate distribution state within the cytoplasm. The cytoplasmic region is divided into rectangular sub-blocks, and the gray-level variation coefficient is calculated to quantify the uniformity of material distribution within the cytoplasm, reflecting the stability of the intracytoplasmic environment. Two-dimensional discrete wavelet multi-scale decomposition is performed to extract the energy ratio between low and high frequencies, revealing texture roughness, and the correlation coefficients of horizontal and vertical detail sub-bands are calculated to quantify the consistency of texture direction.

[0042] A further provision is that the extraction process of the texture feature parameter set in S300 also includes:

[0043] Extract the texture feature vector of the local region of interest in each frame; wherein, the texture feature vector includes: grain intensity, uniformity, frequency domain features and texture direction features;

[0044] Arrange the texture feature vectors in chronological order to form a temporal feature matrix;

[0045] Principal component analysis is performed on the time-series feature matrix to extract the first principal component vector as the dominant mutation direction;

[0046] Project the texture feature vector of each frame onto the dominant mutation direction to obtain a scalar projection value sequence;

[0047] Calculate the mean of the scalar projection value sequence to generate an absolute deviation sequence;

[0048] The standard deviation of the absolute deviation sequence is used as the fluctuation intensity of the cytoplasmic texture;

[0049] The particle intensity, uniformity, frequency domain features, texture direction features, and fluctuation intensity are integrated into the texture feature parameter set.

[0050] By employing the above technical solution, and by extracting the texture feature vectors of the local regions of interest in each frame and arranging the texture feature vectors in chronological order to form a temporal feature matrix, it is possible to assess the changes in texture features of embryos during the thawing and recovery process, providing an important basis for assessing embryo quality. Principal component analysis of the temporal feature matrix can extract the main trends of texture feature changes. By generating an absolute deviation sequence, the fluctuation intensity of texture feature changes can be assessed. The fluctuation intensity can be used to assess the fluctuation intensity of texture feature changes.

[0051] A further setting is that the process of generating the blastocyst formation rate based on the multi-dimensional embryo quality parameters in S400 includes:

[0052] Construct a multimodal parameter fusion input layer: Map the morphological static parameter set, dynamic change parameter set, and texture feature parameter set to a three-dimensional feature space to generate static feature tensors, dynamic feature tensors, and texture feature tensors;

[0053] Establish a three-level cascaded prediction model:

[0054] The first-level sub-model uses a random forest algorithm to perform binary classification pre-screening on the static feature tensor and outputs an embryo structural integrity flag; wherein, the embryo structural integrity flag is a Boolean variable, which is marked as true when the number of effective blastomeres is greater than or equal to a preset first threshold and the proportion of abnormal points in the zona pellucida is less than or equal to a preset second threshold;

[0055] The second-level sub-model uses a long short-term memory neural network to extract the temporal pattern of the dynamic feature tensor and outputs the first score.

[0056] The third-level sub-model uses a support vector regression machine to perform nonlinear fitting on the texture feature tensor and outputs a second score.

[0057] Based on the true / false status of the embryo structural integrity marker, a preset weight allocation strategy is invoked to calculate the blastocyst formation rate.

[0058] By adopting the above technical solutions, multimodal feature tensors are generated through three-dimensional feature space mapping, preserving the spatial correlation and structural information between parameters; a random forest algorithm is used to perform binary classification pre-screening on static feature tensors, and embryonic structural defects are quickly identified based on the number of blastomeres and the integrity of the zona pellucida; a long short-term memory neural network is applied to extract the temporal patterns of dynamic feature tensors, capturing the long-term dependence features of contraction and recovery processes; a support vector regression machine is used to fit the nonlinear relationship of texture feature tensors to analyze developmental potential; and a weight allocation strategy dynamically adjusts the model contribution based on structural integrity, enhancing the evaluation system's tolerance to partially damaged embryos.

[0059] A further setting involves calculating the blastocyst formation rate based on the true / false status of the embryonic structural integrity marker, using a preset weighting strategy.

[0060] If the embryonic structural integrity marker is true, then the blastocyst formation rate satisfies:

[0061] ;

[0062] in, It received the highest rating; It received the second highest score; Blastocyst formation rate; and These are adjustable weighting coefficients;

[0063] If the embryonic structural integrity marker is false, then the blastocyst formation rate satisfies:

[0064] ;

[0065] in, The structural defect compensation factor is determined based on the proportion of abnormal points in the zona pellucida and the number of effective blastomeres. and These are adjustable weighting coefficients.

[0066] By adopting the above technical solution, the assessment model achieves adaptive adjustment based on the weighting strategy of switching between true and false states of embryonic structural integrity markers. When the structure is intact, dynamic scores and texture scores are directly integrated to highlight the dominant role of physiological activity and cytoplasmic state. When there are structural defects, a structural defect compensation factor is introduced, and the predicted value weights are dynamically adjusted according to the proportion of abnormal zona pellucida and the number of blastomeres to objectively reflect the degree of inhibition of developmental potential by structural damage. Adjustable weight coefficients give the model flexible adaptability to different laboratory conditions or embryo types. This mechanism ensures that the assessment results both respect the fundamental role of structural integrity and reasonably quantify the residual developmental potential of defective embryos, thereby improving the precision of clinical decision-making.

[0067] A further setting includes, after the step of calculating the blastocyst formation rate by calling a preset weight allocation strategy based on the true / false status of the embryonic structural integrity marker, the method further includes:

[0068] The blastocyst formation rate is input into the Logistic calibration function to generate a standardized blastocyst formation rate.

[0069] By adopting the above technical solution, the original blastocyst formation rate is probabilistically calibrated using the Logistic calibration function, and the predicted value is mapped to the standardized probability space. This eliminates the systematic deviation between the model output and the actual biological probability. The calibrated standardized probability can be directly compared with the preset threshold range, avoiding the risk of misjudgment due to differences in model dimensions.

[0070] A further setting is that the process in S500 of outputting the level-three embryo quality assessment result based on the comparison between the blastocyst formation rate and a preset threshold range includes:

[0071] Set a preset threshold range: Divide the numerical range corresponding to the standardized blastocyst formation rate into three continuous intervals; wherein, the first interval is defined as greater than or equal to a preset high threshold, the second interval is defined as greater than or equal to a preset low threshold and less than a preset high threshold, and the third interval is defined as less than a preset low threshold.

[0072] Perform a three-tier embryo quality assessment:

[0073] If the standardized blastocyst formation rate is greater than or equal to a preset high threshold, the embryo quality grade is confirmed as excellent.

[0074] If the standardized blastocyst formation rate is greater than or equal to a preset low threshold and less than a preset high threshold, then the embryo quality grade is confirmed as a qualified embryo grade.

[0075] If the standardized blastocyst formation rate is less than a preset low threshold, the embryo quality grade is confirmed as an abnormal embryo grade.

[0076] The embryo quality grade is associated with one of the following grades: excellent embryo grade, qualified embryo grade, or abnormal embryo grade. The corresponding embryo identification code and the standardized blastocyst formation rate are then linked together to output a structured evaluation result that includes the embryo identification code, the standardized blastocyst formation rate, and the embryo quality grade.

[0077] By adopting the above technical solution, the standardized blastocyst formation rate is divided into continuous threshold intervals to avoid information loss caused by discrete grading; by grading into excellent embryo grade, qualified embryo grade and abnormal embryo grade, important references are provided for clinical decision-making; by associating embryo identification code, blastocyst formation rate and quality grade, accurate matching between assessment results and embryo entities is achieved.

[0078] In summary, the present invention has the following beneficial effects: it enables fully automated, multi-dimensional quantitative assessment of the thawing and recovery process of frozen embryos, thereby improving the scientific nature of embryo screening and the clinical pregnancy success rate. Attached Figure Description

[0079] Figure 1 This is the main flowchart of an embodiment;

[0080] Figure 2 This is a flowchart illustrating the extraction process of the morphological static parameter set in S300 of the embodiment;

[0081] Figure 3 This is a flowchart illustrating the extraction process of the dynamically changing parameter set in S300 of the embodiment;

[0082] Figure 4 This is a flowchart illustrating the process of extracting the texture feature parameter set in S300 in the embodiment.

[0083] Figure 5 This is a flowchart illustrating the process of generating blastocyst formation rate based on multi-dimensional embryo quality parameters in step S400 of the embodiment.

[0084] Figure 6 This is a flowchart illustrating the process in S500 of the embodiment where a three-level embryo quality assessment result is output based on the comparison between the blastocyst formation rate and a preset threshold range. Detailed Implementation

[0085] The present invention will be further described in detail below with reference to the accompanying drawings.

[0086] like Figures 1 to 6 As shown;

[0087] This embodiment discloses an image detection and evaluation method for the morphology of thawed fertilized egg embryos, including the following steps:

[0088] S100. Obtain a time-series image sequence of fertilized egg embryos after thawing; wherein, the time-series image sequence includes multiple frames of microscopic images of fertilized egg embryos acquired at fixed time intervals within a predetermined time period in the thawing and resuscitation culture medium; preferably, the predetermined time period is from the 1st minute to the 30th minute;

[0089] Specifically, the microscopic imaging device automatically triggers image acquisition when the fertilized egg / embryo is transferred into the thawed culture medium; then, images are continuously acquired at a predetermined acquisition frequency to generate a time-series image sequence; the predetermined acquisition frequency is set to acquire 6 frames per minute for 30 minutes; finally, embryo identification codes and timestamp tags are added to the time-series image sequence; more specifically, the acquisition device uses an inverted microscope combined with a constant temperature... The incubator and inverted microscope are connected to the automated control module via an API interface. When the fertilized egg embryo is transferred into the thawing and resuscitation culture medium, the liquid level sensor triggers the image acquisition command. The system continuously acquires images at a frequency of 6 frames per minute for 30 minutes, generating a total of 180 time-series image sequences. Each image is associated with an embryo identification code and a timestamp label.

[0090] S200. Perform image standardization processing on the time-series image sequence to generate a standardized embryo image set; wherein, the image standardization processing includes background noise filtering, illumination uniformity correction and image resolution unification operation.

[0091] S300. Extract multi-dimensional embryo quality parameters from a standardized embryo image set; wherein, the multi-dimensional embryo quality parameters include a set of static morphological parameters, a set of dynamic change parameters, and a set of texture feature parameters;

[0092] S400, generating blastocyst formation rate based on multi-dimensional embryo quality parameters;

[0093] S500: Based on the comparison between the blastocyst formation rate and the preset threshold range, output the three-level embryo quality assessment results; among which, the three-level embryo quality assessment results include excellent embryo grade, qualified embryo grade and abnormal embryo grade.

[0094] In one possible implementation, the image normalization process for the time-series image sequence in S200 includes:

[0095] Perform background noise filtering: For each frame of microscopic image in the time-series image sequence, apply a preset Gaussian convolution kernel for spatial domain filtering, and dynamically calculate the noise intensity threshold based on the difference in gray values ​​of adjacent pixels. Pixels below the noise intensity threshold are identified as background noise and filled with morphological closing operations to generate a denoised intermediate image set.

[0096] Preferably, the size of the Gaussian convolution kernel is Pixels and standard deviation The noise intensity threshold satisfies: ;in, Noise intensity threshold; This represents the average grayscale value of a local area. Standard deviation; This is an adjustable coefficient;

[0097] For morphological closing operations, preferably, pixels below the noise intensity threshold are identified as background regions, and a binary noise mask is generated; a circular structuring element with a radius of 3 pixels is used to dilate the binary noise mask, and then the same structuring element is applied to the dilated mask to perform an erosion operation; finally, the connected regions generated by the closing operation are mapped to the gray space of the microscopic image corresponding to the time-series image sequence and mean-filled to generate a denoised intermediate image set.

[0098] Perform illumination uniformity correction: Use the multi-scale Retinex algorithm to calculate the luminance component of each frame of the denoised intermediate image set, and calculate the illumination uniformity index on the luminance component. If the illumination uniformity index is lower than the set threshold, start gamma correction to adjust the global contrast and generate an illumination-equalized image set; otherwise, directly use it as the output as the illumination-equalized image set.

[0099] Preferably, the scale parameter in the multi-scale Retinex algorithm is selected as follows: Pixels, calculating the illumination uniformity index on the luminance component. ;in, This is the global mean of the luminance component; The standard deviation of the luminance component;

[0100] Preferably, the gamma correction adjustment of global contrast process employs a gamma correction function. Execution, among which, The original grayscale value of the input pixel; The maximum grayscale value allowed for the image bit depth; This refers to the corrected grayscale value of the output pixel. For correction index;

[0101] Furthermore, the correction index With light uniformity index Related; it is understandable that when the uniformity of illumination index The lower the value, the smaller the correction index. To significantly improve shadow detail; when the light uniformity index The higher the value, the greater the correction index. To avoid overexposure and damage to highlight details.

[0102] Perform image resolution normalization: Set a baseline resolution, calculate the scaling factor for the illumination-equalized image set, and then use a bicubic interpolation algorithm to scale the image set according to the scaling factor to generate a resolution-normalized image set.

[0103] Preferably, the scaling factor ,in, The width of the source image; The height of the source image; The target width is the baseline resolution; The target height is the baseline resolution. This is the scaling factor; it can be understood that taking... and The smaller value in the range.

[0104] Preferably, the resolution-normalized image set is then converted to a 32-bit floating-point HDR format, and a global histogram specification operation is performed to make its grayscale distribution follow a normal distribution with a mean of 128 and a standard deviation of 40. The image is then encapsulated into a DICOM format sequence file, in which each frame of the image is associated with a corresponding embryo identifier code and a timestamp tag to generate a standardized embryo image set.

[0105] In one possible implementation, the extraction process of the morphological static parameter set in S300 includes:

[0106] Based on each frame of microscopic image from a standardized embryo image set, embryo region segmentation is performed. This segmentation uses a thresholding algorithm combined with edge gradient detection to generate a binarized embryo mask. Next, the geometric features of the region covered by the binarized embryo mask are calculated. These geometric features include:

[0107] Perform connected component labeling on the binarized embryo mask, count the total number of pixels in the largest connected component and multiply it by the single pixel physical size conversion factor to generate the embryo projection area;

[0108] Based on the minimum bounding rectangle of the largest connected region, the ratio of the major axis length to the minor axis length is calculated as the major-minor axis ratio. It can be understood that the larger the ratio, the more elongated the embryo is. This parameter can quickly determine whether the embryo is close to the ideal circle. At the same time, the ratio of the actual perimeter of the largest connected region to the equivalent circle circumference is calculated as the morphological irregularity. It can be understood that the morphological irregularity is used to evaluate the smoothness or complexity of the embryo's outline edge and to detect whether there are irregular changes such as depressions or protrusions.

[0109] A ring-shaped detection area with a width of 5 pixels is constructed around the largest connected component. The number of gray-scale value abrupt change points in this area is calculated and compared with the proportion of the total number of pixels in the ring-shaped area to obtain the proportion of abnormal points in the transparent band. It should be noted that the gray-scale value abrupt change point refers to the gradient magnitude of each pixel in the ring-shaped detection area, which is calculated and compared with a preset threshold. If the gradient magnitude is greater than the preset threshold, the pixel is determined to be a gray-scale value abrupt change point.

[0110] The watershed algorithm is applied to the internal region of the binary embryo mask to segment blastomeres. The number of closed regions with independent centroids is counted to obtain the effective number of blastomeres. It can be understood that after the watershed algorithm is processed, the originally adhered internal region of the embryo is segmented into multiple independent sub-regions. All the segmented sub-regions with independent centroids are traversed and their total number is calculated to obtain the effective number of blastomeres.

[0111] The embryo's projected area, length-to-short axis ratio, morphological irregularity, proportion of abnormal points in the zona pellucida, and number of effective blastomeres are integrated into a set of static morphological parameters.

[0112] Preferably, the embryo region segmentation operation employs a threshold segmentation algorithm combined with edge gradient detection to generate a binary embryo mask, which includes:

[0113] For the current frame microscopic image in the standardized embryo image set, a threshold segmentation algorithm is applied to calculate the segmentation threshold for each pixel, generating an initial segmented binary image; preferably, if the current pixel grayscale value satisfy If the current pixel's grayscale value is... satisfy If this pixel is initially marked as a candidate region for embryonic development, then an initial segmentation binary map is generated; where, This indicates the set threshold; where the pixel value of the embryo candidate region is 255, and the pixel value of the background region is 0.

[0114] Perform edge gradient detection on the initial segmented binary image; including:

[0115] The Sobel operator is applied to the current frame microscopic image of the standardized embryo image set for computation. Directional gradient and The directional gradient is then calculated, and the gradient magnitude of each pixel is obtained.

[0116] Set a gradient magnitude threshold. If the gradient magnitude of a pixel is greater than or equal to the gradient magnitude threshold, and the pixel belongs to the background region in the initial segmentation binary image, then the pixel is reclassified as an embryo candidate region. After traversing each pixel, a binary embryo mask is generated.

[0117] In one possible implementation, the process of extracting the dynamically changing parameter set in S300 includes:

[0118] A temporal alignment operation is performed on a standardized set of embryo images; the temporal alignment operation uses the embryo centroid tracking method for cross-frame position correction.

[0119] Specifically, the process of cross-frame position correction using the embryo centroid tracking method includes:

[0120] In each frame of the image, the center point of the smallest bounding rectangle of the embryonic region is located as the centroid.

[0121] Using the centroid of the starting frame as the origin of the coordinate system, calculate the centroid offset vector for subsequent frames;

[0122] The embryonic regions of all frames are aligned to a unified coordinate system through affine transformation.

[0123] Calculate dynamic change features on the temporally aligned image sequence:

[0124] Within a predetermined time period, a Fast Fourier Transform is applied to the projected area of ​​the embryo in consecutive frames to extract the main peak frequency value with the largest amplitude in the spectrum. This frequency value is then limited to a preset active effective frequency band to obtain the contraction motion frequency. It should be noted that the preset active effective frequency band refers to... ;

[0125] Starting from the time sequence start frame, the contour area change rate is continuously monitored. When the absolute value of the change rate is lower than the preset stable threshold for multiple consecutive windows, the corresponding timestamp is recorded as the recovery delay time. Preferably, the contour area change rate is calculated with a sliding window of 5 frames, and the window is set to 3 consecutive windows.

[0126] Regular grid points were selected within the embryonic cytoplasm region as sampling points. For a continuous time frame sequence, the root mean square error (RMSE) of the grayscale value difference between corresponding sampling points in each pair of adjacent frames was calculated. This RMSE characterizes the overall dynamic change intensity of the cytoplasmic region between adjacent frames; a larger RMSE indicates more drastic changes in the brightness, texture, and structure of the cytoplasm between adjacent frames. These changes typically originate from dynamic processes such as cytoplasmic flow, organelle movement, and changes in molecular concentration. A linear fit was performed on the time series formed by the calculated RMSEs between all adjacent frames to obtain its slope. This slope represents the average rate of change in the intensity of cytoplasmic region changes over time. A slope greater than 0 indicates that the intensity of change is increasing over time, such as more intense cytoplasmic activity; a slope less than 0 indicates that the intensity of change is decreasing over time, such as cytoplasmic activity becoming more gradual or stagnant.

[0127] Preferably, Gaussian filtering is performed on the corresponding sampling points of adjacent frames to denoise them before calculating the root mean square error;

[0128] The contraction frequency, recovery delay time, and slope of change are integrated into a set of dynamically changing parameters.

[0129] In one possible implementation, the extraction process of the texture feature parameter set in S300 includes:

[0130] Local region of interest (ROI) extraction is performed on each frame of a microscopic image from a standardized embryo image set; the ROI extraction is the cytoplasmic region inside the embryo excluding the blastomeres.

[0131] Calculate texture features on the local region of interest; whereby the texture features include:

[0132] In the local region of interest in the embryonic cytoplasm, the rotation-invariant local binary mode operator is applied to calculate its texture spectrum; the cumulative probability of high-frequency modes in the texture spectrum is statistically analyzed, and the cumulative probability value is used as a quantitative index to describe the graininess intensity of the texture in the local region of interest.

[0133] The region of interest in the embryonic cytoplasm is divided into a predetermined number of rectangular sub-blocks. The average gray value of all pixels within each rectangular sub-block is calculated to generate a set of gray value means for each sub-block. Based on this set of gray value means, the standard deviation and arithmetic mean of all sub-block gray value means are calculated to obtain the coefficient of variation (CVA). The CVA serves as a quantitative indicator of the uniformity within the cytoplasm. The CVA is negatively correlated with uniformity; an increase in the CVA indicates a decrease in the spatial uniformity of organelle distribution. It should be noted that the CVA is obtained by dividing the standard deviation by the arithmetic mean.

[0134] Preferably, the preset number of rectangular sub-blocks is set to ;

[0135] Two-dimensional discrete wavelet multi-scale decomposition is performed on the local region of interest; the ratio of low-frequency approximate subband energy to high-frequency detail subband energy is extracted as frequency domain feature, and the Pearson correlation coefficient between horizontal and vertical detail subbands is calculated as texture direction feature.

[0136] It is understandable that frequency domain features ;in, This is the low-frequency approximate subband energy; The energy of the high-frequency detail subband is given; the Pearson correlation coefficient is expressed as... .

[0137] In one possible implementation, the extraction process of the texture feature parameter set in S300 also includes:

[0138] Extract texture feature vectors from the local regions of interest in each frame; where the texture feature vectors include: grain intensity, uniformity, frequency domain features, and texture orientation features;

[0139] The texture feature vectors are arranged in chronological order to form a temporal feature matrix; preferably, Z-score is used for standardization before forming the temporal feature matrix so that the mean of each feature dimension is 0 and the standard deviation is 1.

[0140] Principal component analysis is performed on the time series feature matrix, and the first principal component vector is extracted as the dominant direction of variation.

[0141] Projecting the texture feature vector of each frame onto the dominant mutation direction yields a scalar projection value sequence; the scalar projection value sequence is represented as follows: ,in, Indicates the first The projection value of the frame; Indicates the total number of frames; where, Represents the projected value of the scalar projection value sequence.

[0142] Calculate the mean of the scalar projection value sequence to generate the absolute deviation sequence; the absolute deviation sequence is represented as... ,in, The mean of the scalar projection value sequence; Indicates the first The projection value of the frame; Indicates the total number of frames;

[0143] The standard deviation of the absolute deviation sequence is used as the fluctuation intensity of cytoplasmic texture. It can be understood that the fluctuation intensity quantifies the magnitude of the fluctuation of cytoplasmic texture features over time in their main direction of variation. The higher the index value, the more violent the texture fluctuates over time (the lower the stability); the lower the index value, the smoother the texture fluctuates over time (the higher the stability).

[0144] The intensity of graininess, uniformity, frequency domain features, texture direction features, and fluctuation intensity are integrated into a set of texture feature parameters.

[0145] In one possible implementation, the process of generating blastocyst formation rate based on multidimensional embryo quality parameters in S400 includes:

[0146] Construct a multimodal parameter fusion input layer: Map the set of morphological static parameters, the set of dynamic change parameters, and the set of texture feature parameters to the three-dimensional feature space respectively to generate static feature tensors, dynamic feature tensors, and texture feature tensors;

[0147] Understandably, the static feature tensor is composed of the embryo's projected area, aspect ratio, morphological irregularity, proportion of abnormal points in the zona pellucida, and number of effective blastomeres, linearly combined according to preset weighting coefficients; the dynamic feature tensor is composed of the contraction frequency, recovery delay time, and change slope, which are then spliced ​​together after time series standardization; the texture feature tensor is composed of the graininess intensity, uniformity, frequency domain features, texture direction features, and fluctuation intensity, which are then dimensionality-reduced by principal component analysis to generate a low-dimensional embedding vector.

[0148] Establish a three-level cascaded prediction model:

[0149] The first-level sub-model uses a random forest algorithm to perform binary classification pre-screening on the static feature tensor and outputs an embryo structural integrity marker. The embryo structural integrity marker is a Boolean variable, which is marked as true when the number of effective blastomeres is greater than or equal to a preset first threshold and the proportion of abnormal zona pellucida is less than or equal to a preset second threshold. It can be understood that the number of effective blastomeres reflects the cell division potential, and abnormal zona pellucida affects embryo implantation. Both are indicators of structural integrity. Preferably, the first threshold is set to 4 and the second threshold is set to 5%.

[0150] The second-level sub-model uses a long short-term memory neural network to extract temporal patterns from the dynamic feature tensor and outputs the first score. Specifically, the number of hidden layer neurons in the long short-term memory neural network is set to twice the dimension of the dynamic feature tensor, and the input time step is aligned with the number of frames in the temporal image sequence.

[0151] The third-level sub-model uses a support vector regression machine to perform nonlinear fitting on the texture feature tensor and outputs a second score; specifically, the support vector regression machine uses a radial basis kernel function, and its kernel parameters are determined through grid search optimization.

[0152] Based on the true / false status of embryonic structural integrity markers, a preset weight allocation strategy is applied to calculate the blastocyst formation rate.

[0153] In one possible implementation, the process of calculating the blastocyst formation rate by invoking a preset weight allocation strategy based on the true / false status of the embryo structural integrity marker includes:

[0154] If the embryonic structural integrity marker is true, then the blastocyst formation rate satisfies:

[0155] ;

[0156] in, It received the highest rating; It received the second highest score; Blastocyst formation rate; and These are adjustable weighting coefficients; it should be clarified that... ,and and All are greater than or equal to 0.

[0157] If the embryonic structural integrity marker is false, then the blastocyst formation rate satisfies:

[0158] ;

[0159] in, The structural defect compensation factor is determined based on the proportion of abnormal points in the zona pellucida and the number of effective blastomeres. and These are adjustable weighting coefficients.

[0160] Preferably, the structural defect compensation factor Where 0.6 represents the baseline compensation value, which is a priori value obtained based on regression analysis of historical datasets, indicating the basic developmental potential that the embryo still possesses even if it has structural defects; 0.3 is a set scaling factor, representing the degree of attenuation of the baseline developmental potential. This is a comprehensive defect index. ; This indicates that η is truncated to the upper limit, with a maximum value of 1.5; where, and Based on regression analysis of historical datasets, the values ​​for both were determined to be [values ​​to be determined]. and ; This indicates the percentage of abnormal points in the transparent band; Indicates the effective number of blastomeres.

[0161] In one possible implementation, after calculating the blastocyst formation rate by calling a preset weight allocation strategy based on the true / false states of the embryonic structural integrity marker, the method further includes:

[0162] Input the blastocyst formation rate into the Logistic calibration function to generate the standardized blastocyst formation rate.

[0163] Specifically, the process of generating a standardized blastocyst formation rate includes: ;in, To standardize blastocyst formation rate; This is a parameter for adjusting the slope; This is the decision boundary offset; preferably, and The values ​​are 8 and 0.7 respectively.

[0164] In one possible implementation, the process of outputting the three-level embryo quality assessment result in S500 based on the comparison between the blastocyst formation rate and a preset threshold range includes:

[0165] Set a preset threshold range: Divide the numerical range corresponding to the standardized blastocyst formation rate into three continuous intervals; wherein, the first interval is defined as greater than or equal to a preset high threshold, the second interval is defined as greater than or equal to a preset low threshold and less than a preset high threshold, and the third interval is defined as less than a preset low threshold; preferably, the preset high threshold is set to 0.85 and the preset low threshold is set to 0.65.

[0166] Perform a three-tier embryo quality assessment:

[0167] If the standardized blastocyst formation rate is greater than or equal to the preset high threshold, the embryo quality grade is confirmed as excellent.

[0168] If the standardized blastocyst formation rate is greater than or equal to the preset low threshold and less than the preset high threshold, then the embryo quality grade is confirmed as a qualified embryo grade.

[0169] If the standardized blastocyst formation rate is less than the preset low threshold, the embryo quality grade is confirmed as an abnormal embryo grade.

[0170] By associating one of the following embryo grades—excellent, qualified, or abnormal—with the corresponding embryo identification code and standardized blastocyst formation rate, a structured evaluation data block is generated.

[0171] This specific embodiment is merely an explanation of the present invention and is not intended to limit the invention. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A method for image detection and evaluation of the morphology of thawed fertilized egg embryos, characterized in that, Includes the following steps: S100. Obtain a time-series image sequence of fertilized egg embryos after thawing; wherein, the time-series image sequence includes multiple frames of microscopic images of fertilized egg embryos acquired at fixed time intervals within a predetermined time period in the thawing and resuscitation culture medium; S200. Perform image standardization processing on the time-series image sequence to generate a standardized embryo image set; wherein, the image standardization processing includes background noise filtering, illumination uniformity correction, and image resolution unification. S300. Extract multi-dimensional embryo quality parameters from the standardized embryo image set; wherein, the multi-dimensional embryo quality parameters include a set of static morphological parameters, a set of dynamic change parameters, and a set of texture feature parameters; The extraction process of the dynamically changing parameter set includes: A temporal alignment operation is performed on the standardized embryo image set; wherein, the temporal alignment operation uses the embryo centroid tracking method to perform cross-frame position correction; Calculate dynamic change features on the temporally aligned image sequence: Within a predetermined time period, a fast Fourier transform is applied to the embryo projection area of ​​consecutive frames to extract the main peak frequency value with the largest amplitude in the spectrum, and this frequency value is limited to within a preset effective frequency band to obtain the contraction motion frequency. Starting from the time sequence start frame, continuously monitor the contour area change rate. When the absolute value of the change rate is lower than the preset stable threshold for multiple consecutive windows, record the corresponding timestamp as the recovery delay time. Regular grid points are selected within the embryonic cytoplasm region as sampling points. For a continuous time frame sequence, the root mean square error of the gray value difference between corresponding sampling points between each pair of adjacent frames is calculated. The time series formed by the calculated root mean square errors between all adjacent frames is linearly fitted to obtain its slope. The contraction frequency, recovery delay time, and change slope are combined into the set of dynamic change parameters. S400. Generate blastocyst formation rate based on the multi-dimensional embryo quality parameters; S500. Based on the comparison between the blastocyst formation rate and the preset threshold range, output the three-level embryo quality assessment results; wherein, the three-level embryo quality assessment results include excellent embryo grade, qualified embryo grade and abnormal embryo grade.

2. The image detection and evaluation method for the morphology of thawed fertilized egg embryos according to claim 1, characterized in that: The process of image normalization of the time-series image sequence in S200 includes: Perform background noise filtering operation: For each frame of microscopic image in the time-series image sequence, apply a preset Gaussian convolution kernel for spatial domain filtering, and dynamically calculate the noise intensity threshold based on the difference in gray values ​​of adjacent pixels. Pixels below the noise intensity threshold are identified as background noise and filled with morphological closing operation to generate a denoised intermediate image set. Perform illumination uniformity correction operation: use the multi-scale Retinex algorithm to calculate the luminance component of each frame image in the denoised intermediate image set, and calculate the illumination uniformity index on the luminance component. If the illumination uniformity index is lower than the set threshold, start gamma correction to adjust the global contrast and generate an illumination-equalized image set. Perform image resolution normalization operation: Set a baseline resolution, calculate the scaling factor for the illumination equalization image set, and then use a bicubic interpolation algorithm to scale the image set according to the scaling factor to generate a resolution normalized image set.

3. The image detection and evaluation method for the morphology of thawed fertilized egg embryos according to claim 2, characterized in that: The extraction process of the morphological static parameter set in S300 includes: Based on each frame of microscopic image in the standardized embryo image set, an embryo region segmentation operation is performed; wherein, the embryo region segmentation operation employs a threshold segmentation algorithm combined with edge gradient detection to generate a binary embryo mask; then, the geometric features of the region covered by the binary embryo mask are calculated; wherein, the geometric features include: Connected component labeling is performed on the binarized embryo mask, the total number of pixels in the largest connected component is counted and multiplied by the single pixel physical size conversion factor to generate the embryo projection area; Based on the minimum bounding rectangle of the largest connected region, the ratio of the major axis length to the minor axis length is calculated as the major-minor axis ratio; at the same time, the ratio of the actual perimeter of the largest connected region to the equivalent circle circumference is calculated as the shape irregularity. A ring-shaped detection area is constructed around the largest connected region. The number of gray value abrupt change points in this area and the proportion of the total number of pixels in the ring-shaped area are calculated to obtain the proportion of abnormal points in the transparent band. The watershed algorithm is applied to the internal region of the binarized embryo mask to segment blastomeres, and the number of closed regions with independent centroids is counted to obtain the effective number of blastomeres. The embryo's projected area, aspect ratio, morphological irregularity, proportion of abnormal points in the zona pellucida, and number of effective blastomeres are combined into the set of morphological static parameters.

4. The image detection and evaluation method for the morphology of thawed fertilized egg embryos according to claim 3, characterized in that: The extraction process of the texture feature parameter set in S300 includes: Local region of interest extraction is performed on each frame of the microscopic image of the standardized embryo image set; wherein, the local region of interest is the cytoplasmic region inside the embryo excluding the blastomeres; Calculate texture features on the local region of interest; wherein the texture features include: The texture spectrum of a local region of interest in the embryonic cytoplasm is calculated by applying a rotation-invariant local binary mode operator; the cumulative probability of high-frequency modes in the texture spectrum is statistically analyzed, and the cumulative probability value is used as a quantitative index to describe the graininess intensity of the texture in the local region of interest. The region of interest in the embryonic cytoplasm is divided into a predetermined number of rectangular sub-blocks. The average gray value of all pixels within each rectangular sub-block is calculated to generate a set of gray value averages for each sub-block. Based on the set of gray value averages for each sub-block, the standard deviation and arithmetic mean of the gray value averages of all sub-blocks are calculated to obtain the coefficient of variation. The coefficient of variation serves as a quantitative indicator of the uniformity within the cytoplasm. The coefficient of variation is negatively correlated with uniformity. Two-dimensional discrete wavelet multi-scale decomposition is performed on the local region of interest; the ratio of low-frequency approximate subband energy to high-frequency detail subband energy is extracted as frequency domain feature, and the Pearson correlation coefficient between horizontal and vertical detail subbands is calculated as texture direction feature.

5. The image detection and evaluation method for the morphology of thawed fertilized egg embryos according to claim 4, characterized in that: The extraction process of the texture feature parameter set in S300 also includes: Extract the texture feature vector of the local region of interest in each frame; wherein, the texture feature vector includes: grain intensity, uniformity, frequency domain features and texture direction features; Arrange the texture feature vectors in chronological order to form a temporal feature matrix; Principal component analysis is performed on the time-series feature matrix to extract the first principal component vector as the dominant mutation direction; Project the texture feature vector of each frame onto the dominant mutation direction to obtain a scalar projection value sequence; Calculate the mean of the scalar projection value sequence to generate an absolute deviation sequence; The standard deviation of the absolute deviation sequence is used as the fluctuation intensity of the cytoplasmic texture; The particle intensity, uniformity, frequency domain features, texture direction features, and fluctuation intensity are integrated into the texture feature parameter set.

6. The image detection and evaluation method for the morphology of thawed fertilized egg embryos according to claim 5, characterized in that: The process of generating the blastocyst formation rate based on the multidimensional embryo quality parameters in S400 includes: Construct a multimodal parameter fusion input layer: Map the morphological static parameter set, dynamic change parameter set, and texture feature parameter set to a three-dimensional feature space to generate static feature tensors, dynamic feature tensors, and texture feature tensors; Establish a three-level cascaded prediction model: The first-level sub-model uses a random forest algorithm to perform binary classification pre-screening on the static feature tensor and outputs an embryo structural integrity flag; wherein, the embryo structural integrity flag is a Boolean variable, which is marked as true when the number of effective blastomeres is greater than or equal to a preset first threshold and the proportion of abnormal points in the zona pellucida is less than or equal to a preset second threshold; The second-level sub-model uses a long short-term memory neural network to extract the temporal pattern of the dynamic feature tensor and outputs the first score. The third-level sub-model uses a support vector regression machine to perform nonlinear fitting on the texture feature tensor and outputs a second score. Based on the true / false status of the embryo structural integrity marker, a preset weight allocation strategy is invoked to weight the first score and the second score, and the blastocyst formation rate is calculated.

7. The image detection and evaluation method for the morphology of thawed fertilized egg embryos according to claim 6, characterized in that: The process of calculating the blastocyst formation rate based on the true / false status of the embryo structural integrity markers, using a preset weighting strategy, includes: If the embryonic structural integrity marker is true, then the blastocyst formation rate satisfies: ； in, It received the highest rating; It received the second highest score; Blastocyst formation rate; and These are adjustable weighting coefficients; If the embryonic structural integrity marker is false, then the blastocyst formation rate satisfies: ； in, The structural defect compensation factor is determined based on the proportion of abnormal points in the zona pellucida and the number of effective blastomeres. and These are adjustable weighting coefficients.

8. The image detection and evaluation method for the morphology of thawed fertilized egg embryos according to claim 7, characterized in that: After calculating the blastocyst formation rate by calling a preset weighting strategy based on the true / false status of the embryonic structural integrity marker, the method further includes: The blastocyst formation rate is input into the Logistic calibration function to generate a standardized blastocyst formation rate.

9. The image detection and evaluation method for the morphology of thawed fertilized egg embryos according to claim 8, characterized in that: The process of outputting the level-three embryo quality assessment result in S500 based on the comparison between the blastocyst formation rate and the preset threshold range includes: Set a preset threshold range: Divide the numerical range corresponding to the standardized blastocyst formation rate into three continuous intervals; wherein, the first interval is defined as greater than or equal to a preset high threshold, the second interval is defined as greater than or equal to a preset low threshold and less than a preset high threshold, and the third interval is defined as less than a preset low threshold. Perform a three-tier embryo quality assessment: If the standardized blastocyst formation rate is greater than or equal to a preset high threshold, the embryo quality grade is confirmed as excellent. If the standardized blastocyst formation rate is greater than or equal to a preset low threshold and less than a preset high threshold, then the embryo quality grade is confirmed as a qualified embryo grade. If the standardized blastocyst formation rate is less than a preset low threshold, the embryo quality grade is confirmed as an abnormal embryo grade. The embryo quality grade is associated with one of the following grades: excellent embryo grade, qualified embryo grade, or abnormal embryo grade. The corresponding embryo identification code and the standardized blastocyst formation rate are then linked together to output a structured evaluation result that includes the embryo identification code, the standardized blastocyst formation rate, and the embryo quality grade.

Images