Embryonic Quality Assessment using Advanced Deep Learning Architectures utilizing Microscopic Images of Blastocysts.

The accurate evaluation of embryonic quality which is a key aspect in Assisted Reproductive Technology (ART) and it is crucial to ensure the success of in vitro fertilization (IVF), especially at critical developmental phases like day 3 and day 5. To increase the success rate of implantation, morphological feature-based scoring techniques are currently the mainstay of embryo evaluation. Addressing the need for enhanced accuracy and latency, our work uses advanced algorithms based on deep learning to assess microscopic images of embryos at these critical junctures. The images are pre-processed using histogram equalization and fed to various deep learning models and their performance metrics are compared. Upon comparison, Graph Convolutional Networks achieved the highest accuracy of 96.1% with sensitivity of 96.43% and specificity of 95.24%. Further, the accuracy of the model is increased by implementing an optimized form of GCN known as Graph Attention Networks through an attention mechanism by dynamically determining the importance of each neighbor's features achieving an accuracy of 98.8% with sensitivity of 96.4% and specificity of 97.5%.