Embryo Selection and Classification in IVF Using Hybrid Deep Learning Approach

Abstract

Accurately evaluating and choosing viable embryos for implantation is crucial to the success rate of in vitro fertilisation (IVF).  Traditional evaluation techniques mostly rely on the subjective and variable eye observation of embryologists.  This study suggests a deep learning-based automated embryo classification framework that combines image preprocessing, segmentation, and classification in order to overcome these drawbacks. First, to improve the contrast and clarity of small morphological characteristics in embryo photos, Contrast Limited Adaptive Histogram Equalisation (CLAHE) is used.   The enhanced images are then subjected to a sprint semantic segmentation network (SSS-Net) to ensure that only relevant features support classification and accurately differentiate the embryo region from the backdrop.    Using a hybrid Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) model, the last step successfully classifies embryos into different quality categories. This method captures both spatial and sequential feature dependency. The experimental evaluation demonstrates the durability and dependability of the proposed framework with high precision, recall, F1-score, and accuracy, demonstrating strong predictive performance. When compared to conventional techniques, the combination of sophisticated preprocessing, effective segmentation, and hybrid deep learning architecture greatly increases classification consistency.  All things considered, the suggested method offers an automated, scalable, and objective tool for evaluating the quality of embryos. It may also help embryologists make clinical decisions and eventually increase the success rates of IVF.