3d Vision in Deep Learning Approaches On Assisted Reproductive Technologies of IVF

In vitro fertilization (IVF) is an important assisted reproductive technology (ART) that relies
on accurate embryo assessment and follicle tracking to improve success rates. Recent advances
in deep learning and 3D visualization have provided promising solutions to automate and
improve both embryo and follicle assessment. This study proposes a deep learning framework
for IVF that includes preprocessing, segmentation, and classification techniques. Preprocessing
includes non-local means (NLM) filtering and normalization to reduce noise while preserving
important morphological details in 3D embryo and follicle imaging. This step ensures
improved contrast and clarity, which enables better downstream processing. For segmentation,
a U-Net-based framework is used to precisely define reproductive structures such as oocytes,
embryos, and follicles, which facilitates accurate localization and feature extraction.
Segmentation plays a key role in identifying regions of interest and aiding subsequent
classification. By focusing on the segmented regions, the R-CNN model distinguishes between
viable and non-viable embryos, as well as follicle maturity stages, and automates the grading process with high accuracy. In this approach used as two datasets as 3D ultrasound images and 3D OCT images. The proposed 3D deep learning approach provides an automated, objective, and efficient method for embryo and follicle assessment, which reduces the subjectivity of manual assessment. This study highlights the importance of deep learning-based 3D vision
techniques in revolutionizing IVF procedures and advancing reproductive medicine.