Detecting Blastocyst Components by Artificial Intelligence for Human Embryological Analysis to Improve Success Rate...

Area of Science:

• Reproductive biology and assisted reproductive technologies.
• Medical image analysis and artificial intelligence.
• Computational embryology and deep learning applications.

Background:

• Human blastocyst morphology is critical for in vitro fertilization (IVF) success.
• Current embryologist evaluation of blastocyst components (ZP, TE, BL, ICM) is manual and subjective.
• Deep learning offers potential for automated, objective analysis of blastocyst viability.

Purpose of the Study:

• To propose a novel semantic segmentation network, SSS-Net, for accurate detection of human blastocyst components.
• To enable automated embryological analysis for improved IVF success rates.
• To develop an efficient deep learning model with reduced computational cost.

Main Methods:

• Development of a sprint semantic segmentation network (SSS-Net) utilizing sprint convolutional blocks (SCBs).
• SCBs combine asymmetric kernel and depth-wise separable convolutions for efficiency.
• Implementation of a shallow architecture with dense feature aggregation for enhanced segmentation.
• Evaluation on a public human blastocyst image dataset.

Main Results:

• SSS-Net achieved high Jaccard Index scores for segmenting blastocyst components: ZP (82.88-84.51%), TE (77.40-78.15%), BL (88.39-88.68%), ICM (84.94-84.50%).
• Mean Jaccard Index (Mean JI) reached 85.93% (residual connectivity) and 86.34% (dense connectivity).
• The model demonstrates promising segmentation performance with significantly fewer trainable parameters (4.04 million) than state-of-the-art methods.

Conclusions:

• SSS-Net effectively segments crucial human blastocyst components, aiding automated embryological analysis.
• The proposed network offers an efficient and accurate deep learning solution for IVF embryo assessment.
• This AI-driven approach has the potential to improve IVF success rates through objective embryo selection.