Pool Free Rapid Segmentation Network (PFRS-Net) to detect human blastocyst compartments for embryonic assessment

Assisted reproductive technology has become an increasingly popular solution to address infertility in humans, primarily by in vitro fertilization (IVF). IVF is a complex process where eggs and sperm are combined outside the human body. This occurs in a controlled, specialized laboratory setting that supports and encourages the growth of embryos before they are transferred to the uterus. The IVF process is carefully monitored and regulated within a laboratory environment until the embryos develop and progress to the blastocyst stage. The standard procedure for in vitro fertilization (IVF) involves transferring one or two blastocysts from a batch that has been developed under controlled conditions. A detailed morphological analysis of these blastocysts is performed, assessing their distinct components, including the trophectoderm (TE), zona pellucida (ZP), inner cell mass (ICM), and blastocoel (BL), using manual microscopic techniques. Although deep learning has been successfully utilized in various medical diagnostic and analytical applications, its integration for automating the morphological analysis of human blastocysts continues to face several obstacles. Current methodologies often exhibit inaccuracies and necessitate considerable preprocessing along with expensive computational architectures. As a result, further research is needed to improve the accuracy and efficiency of deep learning techniques in this field to enable their full potential in assisted reproductive technology. To address this challenge, we introduce the Pool Free Rapid Segmentation Network (PFRS-Net), which is specifically developed to effectively identify the compartments of human blastocysts without relying on pooling operations. The network utilizes rapid convolutional block (RCB) modules to achieve accurate detection. The RCB module is specifically designed to capture valuable deep features with computational efficiency. The Swift Decoder block is then used to up-sample the feature maps to their original size using a few layers. This specialized design helps to reduce the number of trainable parameters while maintaining high segmentation accuracy and recovering the lost spatial information using a feature enhancement block (FEB). Our proposed PFRS-Net accurately detects the blastocyst compartments without preprocessing the image and consuming 1.1 million trainable parameters only. This method is trained and tested using a publicly accessible dataset of human blastocyst images. The experimental outcomes demonstrate superior segmentation performance in detecting blastocyst components, which is vital for embryonic research and analysis.