文献速递生殖医学母胎医学

实时人工智能辅助稀有精子搜索:前瞻性多中心配对姊妹卵 pilot 研究

Real-time artificial-intelligence-assisted rare sperm searching: a prospective, multicentre, paired sister-oocyte pilot study.
PubMed 文献 · Reprod Biomed Online · 2026-07-17

该前瞻性多中心配对分皿 pilot 研究纳入 40 例严重男性因素不育患者,比较实时 AI 辅助与无辅助精子搜索。AI 辅助使每皿搜索时间中位数从 35.80 分钟降至 21.80 分钟,每皿回收精子数中位数从 0.00 增至 0.75。小样本探索性分析提示受精率与可用胚胎数可能改善,但需更大样本验证。

要点

  • AI 辅助使每皿搜索时间中位数从 35.80 分钟降至 21.80 分钟(P<0.0001)。
  • AI 辅助使每皿回收精子数中位数从 0.00 增至 0.75(P<0.0001)。
  • 小样本探索性分析提示受精率与可用胚胎数可能改善,需大样本验证。

核心结论

  • 在配对患者内比较中,AI 辅助精子定位减少了搜索时间并增加了精子回收。
  • 胚胎学结果来自小样本探索性分析,需更大规模研究确认。

研究设计与样本

  • 前瞻性、多中心、患者内配对分皿 pilot 研究,时间跨度为 2024 年 1 月至 2025 年 9 月。
  • 纳入 40 例预后不良的非梗阻性无精子症(NOA)或严重少精子症患者,匹配皿含处理后的睾丸组织或精液,分别在有和无 AI 辅助下搜索。

主要发现

  • 每皿搜索时间中位数从 35.80 分钟(IQR 27.85-44.75)降至 21.80 分钟(IQR 18.00-29.88),P<0.0001,基于 40 对配对比较。
  • 每皿回收精子数中位数从 0.00(IQR 0.00-3.50)增至 0.75(IQR 0.00-5.00),P<0.0001。
  • 在两种方法均找到精子的 18 例患者中,每个精子搜索时间中位数从 10.85 分钟(IQR 6.95-38.65)降至 5.50 分钟(IQR 2.27-11.00),P=0.001。
  • 40 例中有 8 例仅使用 AI 辅助识别到的精子进行卵胞浆内单精子注射。
  • 在 12 对均注射两种条件下卵子的患者亚组中,受精率中位数从 0.0%(IQR 0.0-43.8)升至 36.7%(IQR 21.2-63.9),P=0.039;每成熟卵子可用胚胎数从 0.00(IQR 0.00-0.15)增至 0.21(IQR 0.15-0.31),P=0.02。

临床怎么用

  • 对于严重男性因素不育患者,可考虑使用实时 AI 辅助精子搜索以缩短搜索时间并提高精子回收率。
  • AI 辅助可能有助于发现常规搜索遗漏的稀有精子,从而为更多病例提供 ICSI 机会。

局限性

  • 胚胎学结局仅来自小样本亚组(12 对配对患者)的探索性分析。
  • 研究为 pilot 性质,样本量有限,需更大规模研究确认胚胎学获益。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

RESEARCH QUESTION: In severe male factor infertility, does real-time artificial intelligence (AI)-assisted sperm searching reduce search time and increase sperm recovery compared with unassisted searching, and does its use affect downstream embryological outcomes? DESIGN: This was a prospective, multicentre, within-patient paired split-dish pilot study (January 2024-September 2025). Matched dishes containing processed testicular tissue or semen from poor-prognosis non-obstructive azoospermic (NOA) or severe oligozoospermic patients (n = 40) were searched with and without AI assistance. Primary outcomes were time per spermatozoon and time per dish, reported as median (IQR) because distributions were skewed and included zero values. Secondary outcomes were spermatozoa recovered per dish and, in the subset with oocytes injected from both conditions, fertilization and utilizable embryos per mature oocyte injected. RESULTS: Median time per dish decreased from 35.80 min (IQR 27.85-44.75) to 21.80 min (IQR 18.00-29.88) with AI assistance (P < 0.0001; 40 paired comparisons). Median spermatozoa recovered per dish increased from 0.00 (IQR 0.00-3.50) to 0.75 (IQR 0.00-5.00) (P < 0.0001). In 18 patients where spermatozoa were found with both methods, median time per spermatozoon decreased from 10.85 min (IQR 6.95-38.65) to 5.50 min (IQR 2.27-11.00) (P = 0.001). Intracytoplasmic sperm injection proceeded using spermatozoa identified with AI assistance alone in eight of 40 cases. In a small subset of patients with oocytes injected from both conditions (12 paired patients), an exploratory analysis showed that the median fertilization rate increased from 0.0% (IQR 0.0-43.8) to 36.7% (IQR 21.2-63.9) (P = 0.039), and utilizable embryos per metaphase II oocyte increased from 0.00 (IQR 0.00-0.15) to 0.21 (IQR 0.15-0.31) (P = 0.02). CONCLUSIONS: AI-assisted sperm localization reduced search time and increased sperm recovery in paired within-patient comparisons. Embryological findings from a small subset are exploratory and larger studies are required for confirmation.

来源:Reprod Biomed Online,查看原文。版权归原作者所有。

引用

  1. Goss DM, Vasilescu SA, Vasilescu PA, et al. Real-time artificial-intelligence-assisted rare sperm searching: a prospective, multicentre, paired sister-oocyte pilot study. Reprod Biomed Online. 2026. PMID: 42585978. doi:10.1016/j.rbmo.2026.105873

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。