Cleavage-stage embryo segmentation using SAM-based dual branch pipeline: development and evaluation with the...

Area of Science:

• Reproductive Medicine
• Artificial Intelligence
• Medical Imaging

Background:

• Embryo selection is crucial for in vitro fertilization (IVF) success.
• Current embryo assessment is time-consuming, costly, and subjective.
• AI can potentially streamline embryo selection, but existing methods often lack interpretability or focus on later stages.

Purpose of the Study:

• To develop an automated method for segmenting cleavage-stage embryos.
• To improve the accuracy and efficiency of embryo assessment for IVF.
• To address the limitations of current deep learning methods in embryo selection.

Main Methods:

• Introduced a SAM-based dual branch segmentation pipeline.
• Developed the CleavageEmbryo dataset with pixel-level annotations for human cleavage-stage embryos, including fragment information.
• Trained and evaluated state-of-the-art segmentation algorithms on the new dataset.

Main Results:

• The proposed pipeline achieved superior performance in segmenting blastomeres (mAP 0.874) and fragments (Dice 0.695).
• The method demonstrated improved visual quality and accuracy compared to existing algorithms.
• The CleavageEmbryo dataset provides a valuable resource for future research.

Conclusions:

• The SAM-based dual branch segmentation pipeline offers an effective automated solution for cleavage-stage embryo segmentation.
• Accurate segmentation of cleavage-stage embryos can enhance IVF success rates.
• This work provides a foundation for more interpretable and efficient AI-driven embryo selection.