Embryo Development Prediction and Classification System Based on a Double-Layer Optimization Framework

Abstract

The present invention provides an embryo development prediction and classification system based on a double-layer optimization framework, which includes an upper self-supervised training model and a lower embryo development prediction and classification model sharing an encoder; training is carried out using an embryo development image sequence: fixing the model parameters of the lower embryo image development prediction and classification model, and optimizing the model parameters of the upper self-supervised training model; then fixing the model parameters and optimizing the model parameters of the lower embryo image development prediction and classification model; repeating the above steps until the iteration is completed. This collaborative optimization strategy not only improves the accuracy of the model, but also enables it to have stronger adaptability when facing variable and complex medical data.

Description

Technical Field

[0001] The present invention relates to the technical field of embryo development detection, and particularly relates to an embryo development prediction and classification system based on a double-layer optimization framework. Background Art

[0002] During the process of in vitro fertilization (IVF), selecting the most developmentally promising embryo for maternal implantation is one of the key steps to ensure successful pregnancy. Since the developmental potential of each embryo is different, selecting the highest-quality embryo can significantly improve the implantation rate and pregnancy success rate. The quality of the embryo is affected by multiple factors, including the morphological characteristics of the embryo, chromosomal normality, genomic stability, maternal hormone levels, and the preparation of the endometrium. Among them, the morphological assessment mainly focuses on the analysis of embryo images obtained during the embryo development process, which is one of the most widely used methods at present. By comprehensively evaluating these embryo images at different developmental stages, embryos with the highest developmental potential can be selected for transplantation. Therefore, the research on embryo implantation prediction provides a scientific basis for embryologists to select the most suitable embryos.

[0003] Morphology-based embryo assessment generally involves analyzing a series of embryo development images taken by an in-built CCD camera in a time-lapse incubator. These images record the various developmental stages of the embryo from after fertilization to embryo transfer, helping embryologists observe the development of the embryo. By analyzing embryo images, characteristics such as the number of cells, the synchrony of cell division, the symmetry of cell morphology, and the presence of debris in the embryo can be evaluated. Morphological assessment is one of the most commonly used and intuitive embryo screening methods, which can provide a preliminary quality judgment for embryo selection. Traditionally, professionals need to manually conduct systematic assessment and analysis on a large number of images during the embryo development process. This is a very time-consuming task and is prone to human subjective bias. Especially when faced with a large number of embryo images, the efficiency and accuracy of manual assessment often have certain limitations. In addition, some subtle morphological characteristics may be overlooked during manual analysis, although these characteristics are not easily detectable, they may have an important impact on the developmental potential and implantation success rate of the embryo.

[0004] In recent years, with the rise of deep learning algorithms, the evaluation of embryo morphology has been significantly optimized. Deep learning technologies, especially image recognition algorithms such as convolutional neural networks (CNNs), have been widely applied to embryo image analysis. These algorithms can automatically and accurately analyze and evaluate a large number of embryo images, thus improving the efficiency and accuracy of screening. Compared with traditional manual evaluation, deep learning can identify more subtle morphological differences, even those details that are difficult to detect by the naked eye, providing a more reliable basis for embryo selection. By training on embryo images at different developmental stages, deep learning models can gradually learn and identify the key features of high-quality embryos, such as the regularity of cell division, the symmetry of cell morphology, and the degree of fragmentation. These features are often easily interfered by human factors in traditional manual evaluation, while deep learning can extract these potential rules from massive data, achieving a more objective and consistent analysis. The advantage of this technology is that it not only improves the speed and efficiency of evaluation but also provides more refined judgments, helping embryologists make more accurate decisions. At the same time, the introduction of deep learning technology has also solved the problems of "fatigue" and "bias" existing in traditional evaluation methods. When faced with a large number of embryo images, manual evaluation is prone to a decline in attention or evaluation bias due to long-term work, while AI algorithms can maintain a consistent judgment standard at any time, thus reducing human errors.

[0005] For deep learning in the traditional supervised learning manner, its method relies on task-specific label information during the training process. Although it can achieve excellent performance when there is abundant labeled data, this method also has some limitations. First, the limitation of feature learning is a key issue. Supervised learning usually assumes that the features in the data can be directly used to solve the task objective, but this assumption is often too simplistic in practical applications, and feature learning is greatly affected by the data distribution. Especially when the feature space is relatively complex or the task objective is not clear, traditional supervised learning often has difficulty effectively extracting valuable features, resulting in the model being overly dependent on the training data and having poor generalization ability. Second, the training process of traditional supervised learning is easily affected by overfitting. Due to the task-specific objective and loss function only focusing on the labels of the training data during optimization, the model may only perform well on the training set but have weak generalization ability for the test set or unknown data. In the case of large variations in the training data distribution, traditional methods often cannot handle new data well, thus affecting the stability and practicality of the model. In addition, although the combination of self-supervised learning and supervised learning can, to a certain extent, make up for the limitations of the above-mentioned supervised learning, the two are often carried out independently and lack effective synergy. Summary of the Invention

[0006] The present invention proposes an embryo development prediction and classification system based on a double-layer optimization framework to solve the technical problem that in traditional training methods, self-supervised pre-training and the training process of deep learning models are carried out independently, unable to form an effective synergy, resulting in poor prediction accuracy of the trained deep learning model.

[0007] To solve the above technical problem, the present invention provides an embryo development prediction and classification system based on a double-layer optimization framework. The system includes an upper-layer self-supervised training model and a lower-layer embryo development prediction and classification model sharing an encoder.

[0008] The system is trained using an embryo development image sequence: fixing the model parameters of the lower-layer embryo image development prediction and classification model , and optimizing the model parameters of the upper-layer self-supervised training model . Then, fixing the model parameters to optimize the model parameters of the lower-layer embryo image development prediction and classification model; repeating the above steps until the iteration is completed.

[0009] Preferably, the encoder includes a first encoder and a second encoder;

[0010] The first encoder is composed of a module including a 3x3 convolutional neural network and a ReLU activation function;

[0011] The second encoder uses ResNet50 as the basic network.

[0012] Preferably, a bilinear pooling operation is performed on the features extracted by the second encoder.

[0013] Preferably, the first decoder of the upper-layer self-supervised training model is composed of three ConvLSTM modules.

[0014] Preferably, the second decoder of the lower-layer embryo image development prediction and classification model first converts the high-dimensional feature map into a one-dimensional feature vector through a Flatten feature flattening operation, and then performs a non-linear transformation and weighted processing on the flattened features through a KAN module.

[0015] Preferably, the upper-layer self-supervised training model is trained using a point-to-point contrast loss function.

[0016] Preferably, the contrast loss function has the following expression:

[0017] ;

[0018] ;

[0019] ;

[0020] In the formula, represents the sequence of embryo images subsequence the number of samples; represents the total number of feature points in; represents the model prediction the number of samples in the time dimension; and respectively represent the subsequence height and width in the spatial dimension; , both represent indices, represents the subsequence in the index of the th sample, used to represent the index number of another sample in calculating the point-to-point contrast loss; , both are used to represent indices, represents the subsequence in the index number of the th feature point, used to represent the index number of the feature point of another sample in calculating the point-to-point contrast loss; represents the subsequence in the th embryo sample at the true embryo development feature point at the th position; represents the model's prediction of the subsequence in the th embryo sample at the predicted embryo development characteristics at the th position; represents the calculation of and the cosine similarity between; represents the temperature parameter.

[0021] Preferably, the cross-entropy loss function has the following expression:

[0022] ;

[0023] In the formula, refers to the encoded label obtained by one-hot encoding the labels corresponding to a set of embryo image sequences ; refers to the prediction probability of the model for the sequence ;

[0024] Preferably, before training with the embryo development image sequence, the embryo development image sequence is unified in size and reduced.

[0025] The beneficial effects of the present invention at least include: The present invention adopts a double-layer optimization strategy, enabling the fine-tuning stages of the upper self-supervised training model and the lower embryo development prediction and classification model to be co-optimized rather than independently. This co-optimization method avoids the performance bottleneck between pre-training and fine-tuning in traditional methods. Through self-supervised learning, the model can more effectively capture the key features and potential patterns in the embryo development process, helping the model to explore the internal relationships and structural information in the data, enabling the model to better understand the internal feature structure of the data when processing complex medical images, thereby improving the learning efficiency and feature expression ability of the model. In addition, the generalization features learned by the upper self-supervised training model can help the fine-tuning stage of the lower embryo development prediction and classification model adapt to the task faster, enhance the adaptability to new samples, and further enhance the robustness and generalization ability of the model. This co-optimization strategy not only improves the accuracy of the model but also enables it to have stronger adaptability when facing variable and complex medical data. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a schematic diagram of the system structure of an embodiment of the present invention;

[0027] Figure 2 is a schematic diagram of the double-layer optimization process of an embodiment of the present invention;

[0028] Figure 3 is a schematic diagram of the architecture of the network encoder and decoder of an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] The following clearly and completely describes the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0030] Embodiment 1

[0031] As Figure 1 shown, an embodiment of the present invention provides an embryo development prediction and classification system based on a double-layer optimization framework. The system includes an upper self-supervised training model and a lower embryo development prediction and classification model sharing an encoder.

[0032] Specifically, this embodiment mainly focuses on the image data of the entire process of embryo development recorded in a time-lapse incubator to assist embryologists in conducting pre-embryo transfer prediction research. The embryo image data used comes from multiple different reproductive centers, which provides a rich and diverse data source for this embodiment and helps to reflect the accuracy and reliability of the system proposed in this embodiment.

[0033] To ensure the accuracy and representativeness of the image data, the embryo images in all experiments were captured in the incubator through time-lapse photography technology. This technology can continuously record the development process of embryos without disturbing their normal growth and development, thereby obtaining high-quality and high-definition embryo development image data. The dataset covers all key stages of embryo development, from the cell division of fertilized eggs, the differentiation of each embryo cell, to the complete development process at the blastocyst stage. The overall number of embryo images included in this process is between 300 and 500. These images provide rich time-series data for the research, enabling researchers to accurately track the changing laws of embryos at different development stages, and further laying a solid data foundation for the development of automated evaluation and prediction methods based on images. These data not only support the clinical practice in the field of assisted reproduction but also provide valuable reference materials for the application of deep learning algorithms in embryo quality assessment.

[0034] The embryo image dataset collected in this invention was analyzed and evaluated by multiple embryologists, and the principle of the minority obeying the majority was used to distinguish the quality of the image sequences of the embryo development process in the dataset as the label of the model in this invention. The dataset contains a total of 2403 groups of image sequences of the entire process of embryo development. In the experiments of this invention, a ten-fold cross-validation method was used to divide the training set and the test set.

[0035] The system provided in this embodiment includes two parts: an upper self-supervised training model and a lower embryo development prediction and classification model. Among them, the upper self-supervised training model and the lower embryo development prediction and classification model share the same encoder, which provides a basis for collaborative training. In this embodiment, the encoder shared by the two models uses an optimized convolutional module and ResNet50 network structure, including but not limited to LeNet, AlexNet, VGGNet, ResNet series, EfficientNet, as well as Mamba and Transformer models, etc. The decoder of the upper self-supervised training model uses a network architecture stacked with optimized ConvLSTM modules, including but not limited to VAE, GANs, and time-series generation models, etc. The lower embryo development prediction and classification model uses an optimized KAN module, including but not limited to MLP, CNN, and capsule networks, etc.

[0036] The double-layer optimization process is as Figure 2As shown, the parameters of the network model in this embodiment are denoted as ; the parameters involved in the upper self-supervised training model are denoted as ; the parameters involved in the lower embryo development prediction classification model are denoted as . At the initial stage of training, the model is initialized. In the first step, the parameters in the lower embryo development prediction classification model are first fixed, and the parameters in the upper self-supervised training model are optimized to initially obtain the best parameters in the self-supervised task; in the second step, the best parameters in the self-supervised task are fixed and constrained to optimize the parameters in the lower embryo development prediction classification model to obtain the best parameters in the lower embryo development prediction classification model; in the third step, the best parameters in the classification task obtained in the second step are fixed, and the operation of the first step is repeated to optimize the parameters in the upstream self-supervised task; such a cyclic double-layer optimization is performed until the fixed number of iterations is reached, or the performance of the validation set does not improve significantly in several consecutive iterations, such as less than 0.1%, then it can be considered that the model has reached the optimal state, and the optimization is terminated, and finally the best parameters of the model are obtained.

[0037] The double-layer optimization framework adopted by the present invention effectively improves the model performance by combining the advantages of self-supervised learning and supervised learning. In the self-supervised learning stage, the model learns effective feature representations during the embryo development process through unlabeled embryo image data, increasing the generalization ability of the model on this task and enhancing its robustness. Then, in the downstream fine-tuning stage, the labeled embryo classification label data is used to help the model better understand the underlying structure of the data and optimize it. In this way, the feature representation of the embryo development process is significantly improved, and it can perform better in the downstream embryo implantation prediction task. Self-supervised learning enables the model to extract useful information from large-scale unlabeled data, while fine-tuning allows the model to be more precisely adjusted in specific tasks, thereby enhancing the generalization ability and task adaptability of the model. Generally speaking, the double-layer optimization framework not only improves the effects of pre-training and fine-tuning by efficiently combining self-supervised learning and supervised learning, but also enables the model to achieve better performance in environments with scarce data and diverse tasks.

[0038] To further illustrate the effectiveness of the proposed embryo implantation prediction based on the double-layer optimization strategy, the present invention conducts a large number of ablation experiments and comparative experiments.

[0039] In the self-supervised learning pre-training stage, the model learns the deep feature representation of embryo images through unlabeled data. Through a two-layer optimization strategy, the model not only optimizes the initial parameters of the feature extraction network during pre-training but also adjusts the loss function of the downstream classification task. This two-layer optimization process enables the model to gradually improve the effects in the pre-training and fine-tuning stages, effectively capture the key features during embryo development, and thus improve the accuracy of embryo implantation prediction. In the experiments of the present invention, the Adam optimizer is used for parameter optimization, with the learning rate set to 0.005, the batch size to 16, and the number of training epochs to 60. The Adam optimizer combines weight decay and L2 regularization techniques, effectively solving problems such as slow network convergence speed and parameter overfitting, thereby accelerating the model training process and improving the generalization ability.

[0040] Table 1 shows the accuracy of different method models in the embryo implantation prediction task. From the experimental results in the table, it can be found that the system proposed in the present invention is 3.28% higher than the LeNet model in the embryo implantation prediction task, 1.51% higher than the optimal model ResNet152 in the ResNet series networks, and also 2.23% higher than the VGGNet network in terms of accuracy. The results show that the research method based on two-layer optimization proposed in the present invention has better performance.

[0041] Table 1

[0042]

[0043] In summary, the present invention adopts a two-layer optimization strategy, enabling the self-supervised pre-training of the upstream task and the fine-tuning stage of the downstream embryo implantation prediction research to be co-optimized rather than independently. This co-optimization method avoids the performance bottleneck between pre-training and fine-tuning in traditional methods. Through self-supervised learning, the model can more effectively capture the key features and potential patterns during embryo development, helping the model to explore the internal relationships and structural information in the data, enabling the model to better understand the internal feature structure of the data when processing complex medical images, thereby improving the learning efficiency and feature expression ability of the model. In addition, the generalized features learned in the upstream self-supervised pre-training stage can help the fine-tuning stage of the downstream embryo implantation prediction task to adapt to the task faster, improve the adaptability to new samples, and further enhance the robustness and generalization ability of the model. This co-optimization strategy not only improves the accuracy of the model but also enables it to have stronger adaptability when facing variable and complex medical data.

[0044] Example 2

[0045] Based on Example 1, this example improves the encoders and decoders of the upper-layer self-supervised training model and the lower-layer embryo development prediction classification model.

[0046] Specifically, the encoder shared by the upper - layer self - supervised training model and the lower - layer embryo development prediction and classification model includes a first encoder and a second encoder, and its structure is as Figure 3 shown.

[0047] In this embodiment, the design of the encoder aims to fully extract and model the real development features in embryo images, providing accurate feature representations for subsequent prediction and analysis tasks. The first encoder is mainly composed of a module containing a 3x3 convolutional neural network and a ReLU activation function. Its core goal is to extract real development features from embryo images and serve as the supervision source for predicting embryo development features in the self - supervised task. Through the operation of small convolutional kernels, this encoder accurately extracts local features and, combined with the ReLU activation function, enhances the non - linear expression ability of the model, thereby strengthening the ability to identify and model embryo development features.

[0048] As the backbone structure of the network, the second encoder is not only the shared part in classification tasks and self-supervised tasks but also the core feature extraction module of the model. The second encoder first uses ResNet50 as the basic network to extract local and global features in embryo images with its powerful representation ability. Through multi-level convolutions and residual connections, ResNet50 can effectively capture the deep semantic information in the image, thus generating a feature map containing multi-level embryo development information. This feature map not only retains the detailed features of the image spatially but also encodes rich high-order semantic information in the channel dimension, providing a solid foundation for subsequent fine-grained feature analysis and task modeling. After feature extraction, the second encoder further introduces the Bilinear Pooling operation to capture the high-order interaction features between different local regions in embryo images. Bilinear Pooling generates high-order feature representations through outer product calculations, which can effectively model the complex relationships between local regions, thereby enhancing the model's ability to express fine-grained features. This method not only improves the model's sensitivity to local details but also captures the subtle differences between different components or regions, which is crucial for distinguishing different stage information during embryo development. In addition, in the Bilinear Pooling operation, the second encoder generates a global feature map and an attention map in parallel. The global feature map is extracted by ResNet50, providing rich overall semantic information; the attention map is generated through convolutional layers, reflecting the importance of each region in the image. This attention mechanism can guide the network to focus on the key regions related to embryo development and implantation prediction. In subsequent processing, the feature map and the attention map are combined to further optimize the model's feature representation by weighting the feature map. This combination enables the network to automatically identify and strengthen the attention to key information during embryo development while suppressing the influence of irrelevant or interfering regions. This not only improves the network's ability to capture fine-grained features but also provides stronger support for downstream tasks such as predicting the success rate of embryo implantation. Overall, through the multi-level feature extraction of ResNet50, the high-order feature modeling of Bilinear Pooling, and the introduction of the attention mechanism, the second encoder deeply represents embryo images in both spatial and semantic dimensions, laying the foundation for the excellent performance of the entire model in embryo development stage classification and prediction.

[0049] In this embodiment, the decoder is designed to achieve the prediction of future development stages and the classification of the embryo development prediction stage by making full use of the embryo development feature information extracted by the encoder. The task of the first decoder of the upper-level self-supervised training model is to extract useful information from the spatio-temporal feature representation obtained from the encoder to predict the future stages in the embryo development process. To this end, this embodiment adopts an encoder architecture composed of three ConvLSTM modules. Each ConvLSTM module combines convolutional operations and the LSTM structure, and can learn spatial features and temporal dependencies simultaneously when processing time-series data. These three consecutive ConvLSTM modules extract higher-level spatio-temporal features layer by layer, so that the finally output feature representation contains rich spatio-temporal information, which can provide sufficient support for subsequent predictions. In this way, the first decoder can not only capture the fine-grained features of the current embryo development stage, but also effectively perform temporal modeling to predict the future development process of the embryo.

[0050] The second decoder of the lower-level embryo image development prediction and classification model focuses on predicting and classifying using the embryo development feature information extracted by the encoder, and finally completes the classification task of whether different embryo development processes are suitable for further implantation into the mother. The workflow of the second decoder first converts the high-dimensional feature map into a one-dimensional feature vector through the Flatten feature flattening operation, so that the subsequent classification task can be processed in the simplified feature space. This operation integrates the features of the spatial and temporal dimensions into a compact one-dimensional vector, providing an effective input for the classification model. Next, the second decoder performs nonlinear transformation and weighting on these flattened features through the KAN (Key Attention Network) module. The KAN module uses a learnable edge activation function, which breaks through the limitation of the fixed activation function in the traditional multi-layer perceptron (MLP), so that the activation function of each node can be adaptively adjusted according to the input features, thereby effectively enhancing the network's adaptability to complex embryo development patterns. By introducing this flexible activation mechanism, the KAN module enables the decoder to more accurately learn the features that are critical to the classification task and automatically focus on the most distinguishing areas in the embryonic development process. This design enables the second decoder to not only complete complex multi-class classification tasks, but also show higher accuracy and robustness when dealing with fine-grained classification. In general, the two decoders are designed according to different tasks by introducing multi-layer ConvLSTM modules, Flatten feature flattening operations, and learnable edge activation functions of the KAN module, so that the model can achieve excellent performance in both time series prediction and classification tasks. The first decoder effectively uses the spatiotemporal features extracted by the encoder to predict future developmental stages, while the second decoder achieves accurate classification of embryonic development stages through efficient feature selection and weighting mechanisms. Through this design that combines time series modeling with classification tasks, the model can better understand and predict complex patterns in embryonic development.

[0051] Example 3

[0052] Based on Example 1, this example constructs a loss function for the training process of the upper self-supervised training model.

[0053] Specifically, the goal of the loss function in the upper self-supervised training model is to predict the effective embryonic development feature representation of the next subsequence stage based on the current embryonic image subsequence. This process uses the developmental features of different subsequences in the embryonic image as the supervision source of the self-supervised learning task, and does not need to rely on artificial labels. Therefore, the inner loss function used in this embodiment is a point-to-point contrast loss function. :

[0054] ;

[0055] ;

[0056] Among them, represents the number of embryo image samples in the current subsequence; in the subsequence represents the subsequence the total number of feature points in, and its calculation method is: , represents the predicted subsequence the number of samples in the time dimension, and respectively represent the height and width of the predicted subsequence in the spatial dimension; , representing two indices for traversing all embryo image samples in the embryo subsequence, represents the currently considered embryo image sample, used to represent the index of another sample in calculating the point-to-point contrast loss. When , it means they are the same sample. When , it means they are not the same sample, and their corresponding feature points belong to the negative sample pair; , representing two indices for traversing the feature points of all embryo image samples in the embryo subsequence, represents the index of the feature point of the currently considered embryo image sample, used to represent the index of the feature point of another sample in calculating the point-to-point contrast loss. When calculating the loss of each feature point, and will jointly determine with which feature points are positive sample pairs and which are negative sample pairs; represents the true embryo development feature at the th position in the th embryo sample in the subsequence ; represents the predicted embryo development feature at the th position in the th embryo sample by the model; represents calculating and the cosine similarity between; represents the temperature parameter, which is used to adjust the scale of the similarity and affect the sensitivity of the point-to-point contrast loss function. By adopting the point-to-point contrast loss function, the model can effectively capture the development features of embryos at different time points and improve the model's ability to distinguish between good and bad embryo development processes.

[0057] Example 4

[0058] Based on Embodiment 1, this embodiment constructs a loss function for the training process of the lower-layer embryo development prediction classification model.

[0059] The goal of the loss function of the lower-layer embryo development prediction classification model is to fine-tune the model so that it can better complete specific downstream classification tasks. The cross-entropy loss function is favored due to its wide applicability and various advantages in classification problems. In this embodiment, this function maximizes the log-likelihood in the embryo images by minimizing the loss value, thereby improving the accuracy of the model in predicting the quality of the embryo development process. It has high sensitivity, can quickly respond to changes in the model performance, and accelerate the convergence process of the model. In addition, the cross-entropy loss function is easy to calculate, has good numerical stability and probability interpretability, which makes it the preferred loss function in this embodiment. , and its calculation formula is as follows:

[0060] ;

[0061] where, refers to a set of embryo images whose corresponding labels are encoded labels obtained through a one-hot encoder, refers to a set of embryo images and the predicted probabilities obtained through the network model in the supervised task part.

[0062] Embodiment 5

[0063] Based on Embodiment 1, this embodiment preprocesses the embryo image sequence.

[0064] The original size of the embryo image data collected by the present invention is 500x500 pixels. To better meet the input requirements of the deep learning model, the size of each image is reduced to 224x224. This adjustment ensures that the images conform to the standard input size of the neural network. Especially when using a pre-trained model, the input is usually required to be 224x224 pixels. In addition, reducing the image size helps to reduce the consumption of computing resources, speeds up model training and inference, and thus improves the overall computing efficiency. The smaller input images reduce memory occupancy and computational overhead, effectively improving the training efficiency without sacrificing key information. During the adjustment process, a scaling technique that maintains the aspect ratio of the image is adopted to ensure that the core information of the embryo image is not lost. Through these processes, the model can focus on the key areas of the image, avoid interference from irrelevant backgrounds, and thus improve the classification accuracy. At the same time, the unified image size also simplifies the subsequent data processing and batch processing. Especially when dealing with large-scale data sets, standardizing the size to 224x224 can optimize model training, reduce the complexity during training, and improve the generalization ability of the model. Through these preprocessing operations, this embodiment can not only effectively improve the accuracy of embryo image classification, but also significantly reduce the computational cost, providing more efficient and stable support for subsequent intelligent diagnosis and medical decision-making.

[0065] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. Only the preferred embodiments of the present invention are expressed. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the present invention. As long as the combination of these technical features does not conflict, it should be considered as the scope recorded in this specification.

[0066] It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the appended claims.

Claims

1. An embryo development prediction and classification system based on a double-layer optimization framework, characterized in that: The system includes an upper self-supervised training model and a lower embryo development prediction and classification model, and an encoder is shared between the upper self-supervised training model and the lower embryo development prediction and classification model; Train the system using an embryonic development image sequence: Fix the model parameters of the lower-layer embryonic development prediction classification model , and optimize the model parameters of the upper-layer self-supervised training model ; then fix the model parameters and optimize the model parameters of the lower-layer embryonic development prediction classification model; repeat the above steps until the iteration is completed; The encoder includes a first encoder and a second encoder; The first encoder is composed of a module containing a 3x3 convolutional neural network and a ReLU activation function; The second encoder uses ResNet50 as the basic network; A bilinear pooling operation is performed on the features extracted by the second encoder; The first decoder of the upper self-supervised training model consists of three ConvLSTM modules; The second decoder of the lower embryo development prediction and classification model first converts the high-dimensional feature map into a one-dimensional feature vector through a Flatten feature flattening operation, and then performs a non-linear transformation and weighting process on the flattened features through a KAN module.

2. The embryo development prediction and classification system based on a double-layer optimization framework according to claim 1, wherein: The upper self-supervised training model is trained using a point-to-point contrast loss function.

3. The embryo development prediction and classification system based on a double-layer optimization framework according to claim 2, wherein: The contrastive loss function has the following expression: ； ； ； In the formula, represents the embryonic image sequence subsequence the number of samples; represents the total number of feature points in; represents the model prediction the number of samples in the time dimension; and respectively represent the subsequence height and width in the spatial dimension; , both representing indices, represents the subsequence in the index of the th sample, used to represent the index number of another sample in calculating the point-to-point contrast loss; , both used to represent indices, represents the subsequence in the index number of the th feature point, used to represent the index number of the feature point of another sample in calculating the point-to-point contrast loss; represents the subsequence in the th embryonic sample at the th position of the true embryonic development feature point; represents the model's prediction for the subsequence in the th embryonic sample at the th position of the predicted embryonic development feature; represents the calculation of and the cosine similarity between; represents the temperature parameter.

4. A classification system for predicting embryonic development based on a double-layer optimization framework according to claim 1, characterized in that: The lower embryo development prediction and classification model is optimized using a cross-entropy loss function.

5. The embryo development prediction and classification system based on a double-layer optimization framework according to claim 4, characterized in that: The cross-entropy loss function has the following expression: ； In the formula, represents a set of embryo image sequences The corresponding labels are encoded labels obtained by a one-hot encoder; represents The predicted probability obtained by the upper self-supervised training model.

6. The embryo development prediction and classification system based on a double-layer optimization framework according to claim 1, characterized in that: Before training with the embryo development image sequence, the embryo development image sequence is unified in size and scaled down.

Images