A new deep-learning model using YOLOv3 to support sperm selection during intracytoplasmic sperm injection procedure

Area of Science:

• Reproductive Medicine
• Artificial Intelligence
• Sperm Analysis

Background:

• Sperm morphology and motility are critical indicators of male fertility.
• Accurate and efficient sperm analysis is essential for assisted reproductive technologies (ART).
• Current methods for sperm evaluation can be time-consuming and labor-intensive.

Purpose of the Study:

• To develop and validate a machine learning model using YOLOv3 for simultaneous sperm morphological assessment and tracking.
• To adapt the model for real-time video analysis using an inverted microscope.
• To improve the efficiency of sperm data acquisition and annotation.

Main Methods:

• An artificial intelligence (AI) model based on YOLOv3 was created.
• The model was trained and evaluated using video data from Japanese patients undergoing intracytoplasmic sperm injection (ICSI).
• Performance metrics included sensitivity and positive predictive value (PPV) for morphological assessment and tracking accuracy.

Main Results:

• The AI model demonstrated high sensitivity (0.881) and PPV (0.853) for abnormal sperm morphology.
• For normal sperm, sensitivity was 0.794 and PPV was 0.689.
• Tracking performance showed that 78.4% of sperm objects were mostly tracked, with none completely lost.

Conclusions:

• Training YOLOv3 enables a single model to evaluate sperm morphology and track sperm simultaneously.
• The developed model can acquire time-series data of individual sperm.
• This AI approach can significantly assist in the acquisition and annotation of sperm image data for research and clinical applications.