精液分析の機械学習評価により新たな不妊関連マーカーが明らかになる可能性：パイロットスタディ

科学分野:

• andrology
• reproductive medicine
• machine learning in healthcare

背景:

• Male infertility affects a significant portion of couples seeking fertility treatment.
• Current diagnostic work-up for male infertility can be complex and may benefit from advanced analytical tools.
• Semen analysis is a cornerstone of male infertility evaluation, but its interpretation can be challenging.

研究 の 目的:

• To pilot the application of machine learning models for semen analysis evaluation.
• To assess the potential of machine learning in improving the diagnostic process for male partners of infertile couples.
• To identify key variables influencing semen analysis outcomes using machine learning.

主な方法:

• Retrospective observational study using real-world data from two Italian tertiary centers.
• Utilized two distinct datasets (UNIROMA and UNIMORE) with varying combinations of semen analysis, hormonal, biochemical, ultrasound, and environmental pollution data.
• Applied XGBoost machine learning analysis separately to each dataset.

主要な成果:

• The UNIROMA dataset (n=2,334) showed high accuracy (AUC=0.987) in predicting azoospermia, with follicle-stimulating hormone, inhibin B, and testicular volume as key predictors.
• The UNIMORE dataset (n=11,981) demonstrated good predictive accuracy (AUC=0.668) for azoospermia, highlighting environmental pollution (PM10, NO₂) and biochemical data (white blood cells, red blood cells) as crucial variables.
• Machine learning effectively identified influential predictive variables in both datasets.

結論:

• Machine learning models can be valuable tools in the diagnostic work-up of male infertility.
• Semen analysis results appear to be interconnected with testicular ultrasound characteristics, biochemical markers, and environmental pollution.
• This pilot study supports further investigation into AI-driven approaches for male reproductive health assessment.