NEUROVISION EMBRYONET: A HYBRID DEEP LEARNING FRAMEWORK FOR HIGH-PRECISION EMBRYO MORPHOLOGY ANALYSIS

Abstract

Embryo classification in India has evolved significantly within assisted reproductive technology 
(ART), particularly for in vitro fertilization (IVF). IVF centers in India have grown rapidly, with over 
1,500 clinics by 2023, performing around 2.5 lakh cycles annually, as per the Indian Society of 
Assisted Reproduction. The objective of embryo classification using AI, specifically CNN with CBC, 
is to accurately categorize embryos as "normal" or "viable" for better IVF success. It means 
automating and objectifying the selection process, reducing human error, and enhancing implantation 
potential through advanced image analysis. Manual embryo classification involves embryologists 
visually assessing embryo morphology under a microscope, typically using the Gardner grading 
system. They evaluate features like blastocyst expansion, inner cell mass, and trophectoderm quality 
at specific stages. Embryos are scored and ranked for transfer or cryopreservation, relying on the 
embryologist’s expertise to identify the most viable ones for implantation.  Manual embryo 
classification suffers from subjectivity, as embryologists’ assessments vary due to experience, fatigue, 
or inconsistent criteria application, leading to inter- and intra-observer variability. It’s time
consuming, often missing subtle morphological details, and fails to predict implantation accurately, 
with success rates stagnating at 30-40%. Additionally, manual methods cannot process large datasets 
or integrate clinical variables effectively. The motivation for using AI, CNN, and CBC in embryo 
classification stems from overcoming manual system limitations like subjectivity and low accuracy. 
AI improves consistency, achieving near-perfect accuracy (e.g., 99.06% in CNN with CBC), enhances 
prediction of implantation potential, and reduces human error. It enables faster, data-driven decisions, 
addressing the scalability issues and variability in traditional methods. The proposed AI system, using 
a CNN with CBC model, revolutionizes embryo classification by automating and enhancing accuracy 
in IVF. The CNN extracts detailed features from microscopic embryo images, resizing them pixels 
and normalizing pixel values. These features are then classified by the CatBoost Classifier, achieving 
good accuracy. This hybrid model minimizes misclassification, ensures objective selection, and 
significantly boosts implantation success rates, outperforming manual methods.