1. Development of Segmentation Model for Evaluation of Denuded MII Oocytes
2. Development of Mask Model for Prediction of Blastocyst Development
3. Comparison of Ensemble Model Incorporating DL for Oocyte Evaluation
4. Subgroup Analysis and External Validation of Oocyte Evaluation Models

In a groundbreaking development in the field of assisted reproductive technology, researchers have introduced two innovative models to evaluate static images of denuded MII oocytes. These models aim to predict whether an oocyte will develop into a blastocyst, a crucial step in the process of in vitro fertilization (IVF).

The first model focuses on multiclass segmentation of the oocyte into ooplasm, ZP, and PVS, creating masks that serve as inputs for the second model—a classifier model known as the mask model. This classifier model extracts features from the masks to generate predictions regarding blastocyst development.

To develop the segmentation model, a dataset of 7792 denuded MII oocytes was utilized, with images obtained from fertility clinics globally as well as from an open-source dataset. The model, based on the Fully Convolutional Branch TransFormer (FCBFormer) architecture, demonstrated high accuracy in segmenting the oocyte into its various regions.

The mask model, on the other hand, was developed using a dataset of 51,831 static images of denuded MII oocytes with known blastocyst development outcomes. Features extracted from the oocyte images, along with patient characteristics, were used as inputs for the model to predict blastocyst formation.

Two approaches were explored for oocyte classification: the Auto-sklearn ensemble model and the LightGBM framework. Additionally, a ConvFormer DL model was incorporated into an ensemble model with the LightGBM model to assess the benefits of combining deep learning and machine learning approaches.

Subgroup analyses by clinic and age group were conducted to evaluate the clinical relevance of the models for different patient populations. Furthermore, external validation of the models was performed on a dataset from a Spanish clinic to assess their performance in a real-world setting.

Overall, these models represent a significant advancement in the field of reproductive medicine, offering a more accurate and efficient way to predict blastocyst development in IVF cycles. The integration of deep learning and machine learning techniques holds promise for improving outcomes in assisted reproduction and enhancing patient care.