A deep learning method and system for multi-stage classification of embryonic development

Abstract

The application discloses a deep learning method and system for embryo development multi-stage classification, belongs to the embryo development multi-stage classification field, carries out blocking operation on the preprocessed embryo image, and inputs the same to a double-branch local feature fusion module to carry out local feature extraction, obtains a feature map containing fine-grained local features; the feature map containing the fine-grained local features is input to a deep neural network module to carry out global feature extraction and coding, and a final feature map is obtained; the final feature map is input to a classifier to carry out development stage prediction, and the predicted embryo development stage is output. The advantages of the double-branch local feature fusion module and the deep neural network module are combined for embryo development stage classification, which can not only fully extract the local and global information of the embryo image, improve the accuracy of the classification stage, but also provide a new idea for the application of deep learning in IVF.

Images