Research And Implementation Of Embryo Evaluation System Based On Deep Learning

With the improvement of modern medical level and the deepening of research on the mechanism of embryo development,in vitro fertilization-embryo transfer(IVF-ET)technology has become more mature.Among them,the quality assessment before embryo implantation is an important part of assisted reproductive technology(ART),and is closely related to the success rate of treatment.Morphological evaluation is one of the most widely used evaluation methods.Based on the traditional morphological method,this thesis combines deep learning techniques to design a deep convolutional neural network to automatically identify the main morphological features from embryo images,and the new embryo features are constructed to train the evaluation model,resulting in an efficient,accurate,user-friendly embryo evaluation system.The main research contents of this thesis are as follows:1.Research on the embryonic body segmentation model based on U-Net and residual structure.Due to the large amount of background interference in the original embryo image,it is necessary to remove the area outside the embryo body.Based on the encoderdecoder architecture of U-Net semantic segmentation model,and combining the residual structure,the embryo segmentation network is designed to improve the representation ability of the network.At the same time,in order to improve the speed of the model,the depthwise separable convolution is applied to the design of the residual structure,which greatly improves the segmentation speed.2.Research on the embryological morphological feature recognition model based on multi-task learning.On the basis of embryo segmentation,the multi-task learning strategy is applied to achieve the recognition of the main morphological features simultaneously.This thesis builds a shared weight layer based on ShuffleNet V2 network,and uses the pre-training weight of ImageNet dataset to design a classifier for different morphological features.The multi-task loss based on adaptive weight is adopted to train the model to achieve higher accuracy.3.Research on the embryo evaluation model that combines developmental trends and mean developmental levels.Since embryonic development is a dynamic process and there is an internal priority for all embryos to be selected in the same patient,this thesis designs the embryo development trend and average development based on the main morphological characteristics of the embryo and the expert experience in the field of embryo evaluation.This two new types of features taking into account the dynamic global characteristics of the embryo,then predicting the embryo quality score based on the support vector machines(SVM)model.4.Design and implement a Qt5-based embryo quality assessment system.Combined with the actual application requirements of embryo evaluation,this thesis analyzes the functional module structure of the software,then uses Qt5 to design and implement the graphical user interface and interaction logic.The core algorithm is implemented based on Python,and uses MySql as the database to store evaluation result.The system provides the embryo evaluation core function to users through a graphical interface,which meets the actual work needs.This thesis verifies the integrity and availability of the system by testing system functionality and performance.The research results help to improve the efficiency of embryo evaluation and have good application significance.