Related Experiment Video
Updated: Jul 15, 2026
05:43
Capturing Cytoskeleton-Based Agitation of the Mouse Oocyte Nucleus Across Spatial Scales
Published on: January 12, 2024
Automated Cytoplasmic Image Analysis System for Non-invasive Oocyte Quality Screening Workflow Using Static
Takashi Morimoto1,2, Hidehiko Matsubayashi3, Takumi Takeuchi3
1Graduate School of Information Science, University of Hyogo, 7-1-28, Minatojimaminami-Machi, Chuo-Ku, Kobe, 650-0047, Japan. tmorimoto0619@gmail.com.
Reproductive Sciences (Thousand Oaks, Calif.)
|July 13, 2026
Summary
A deep neural network (ResNet50) can non-invasively assess oocyte quality and predict fertilization using cytoplasmic features from bright-field images. This AI approach offers a new method for evaluating oocyte maturity and developmental potential.
Area of Science:
- Reproductive Medicine and Assisted Reproductive Technology (ART)
- Artificial Intelligence (AI) in Healthcare
- Biomedical Image Analysis
Background:
- Oocyte quality assessment is crucial for successful in vitro fertilization (IVF) outcomes.
- Traditional methods for evaluating oocyte maturity and developmental potential are often subjective and time-consuming.
- There is a need for objective, non-invasive methods to predict embryonic development from oocyte morphology.
Purpose of the Study:
- To evaluate the efficacy of a pre-trained deep neural network (ResNet50) in assessing oocyte quality.
- To determine if ResNet50 can predict subsequent embryonic development by analyzing cytoplasmic features from static bright-field images.
- To assess the non-invasive capabilities of AI in oocyte evaluation before intracytoplasmic sperm injection (ICSI).
Main Methods:
- A prospective cohort study included 306 metaphase II oocytes from 63 patients.
- Static bright-field images were acquired before piezo-ICSI, and cytoplasmic regions were segmented.
- A pre-trained ResNet50 model extracted features, reduced via principal component analysis (PCA), and analyzed using logistic regression and support vector machines (SVM).
Main Results:
- The ResNet50-based system achieved an area under the curve (AUC) of 0.679 for spindle detection and 0.521 for fertilization prediction using logistic regression with PCA.
- Single-feature analysis showed high specificity for spindle detection (AUC=0.656) and moderate specificity for fertilization (AUC=0.607).
- Blastocyst prediction performance was limited (AUC ≤ 0.64), but spindle-negative oocytes showed better fertilization prediction (AUC=0.835).
Conclusions:
- A deep neural network can quantitatively and non-invasively assess oocyte maturity and predict fertilization using cytoplasmic features from static images.
- This AI-driven approach does not require large, specific training datasets for oocyte evaluation.
- The system shows promise for improving objective oocyte quality assessment and predicting fertilization outcomes in ART.
