【作者】 何俊；

【导师】 钟振声；

【作者基本信息】 安徽医科大学 ， 细胞生物学， 2025， 硕士

【摘要】 男性不育是导致不孕不育的主要因素之一,精子质量直接影响受精和胚胎发育的成功率。目前,精子质量评估大多依赖单一的评估指标,如精子活力、形态学、DNA完整性和顶体反应能力,但现有的检测方法无法全面准确地反映精子质量,且存在主观性和误差。为了解决这些问题,本研究提出了一种结合光镊与深度学习技术的新型多参数检测方法,能够实现对单个精子细胞的全面、无创、无标记检测。通过光镊技术对单个精子进行精准捕获,并结合深度学习模型,进行精子活力、体积、表面积以及DNA碎片化等参数的多维度分析。本研究首先通过光镊捕获单个精子细胞,并分析其头部的绕长轴旋转频率和停滞时长占比,以评估精子的活力。与传统的精子活力测量方法相比,光镊技术能够实现对同一个精子不同时刻的多次活力评估。此外,结合深度学习技术,使用U-Net网络对精子头部进行分割,实现了自由游动及光镊捕获的精子头部的三维表面重建,从而提供了更为精准的体积和表面积测量。这一技术避免了传统方法中基于椭球体或球体假设的误差,能够更为准确地测量精子头部的实际三维形态。进一步地,研究还通过生成对抗网络,开发了精子DNA碎片化的虚拟染色方法。该方法不需要使用荧光染料,避免了传统染色方法对精子活力的影响,通过训练深度学习模型,结合明场图像生成虚拟染色图像,实现了对精子DNA碎片化程度的无标记检测。通过虚拟染色技术,为精子DNA碎片化检测提供了新的解决方案,未来希望结合光镊技术,实现对精子活力检测的同时进行DNA碎片化程度的评估。通过将光镊与深度学习技术相结合,本研究实现了在明场照明条件下单精子细胞的多参数检测,为精子质量的全面评估提供了强有力的技术支持。这一方法不仅能够应用于卵胞浆内单精子显微注射技术等辅助生殖技术中的单精子质量筛选,提升后续胚胎发育的成功率,同时也为男性不育的诊断提供了新的数据驱动的解决方案。未来,随着光镊和深度学习技术的进一步优化和完善,该方法有望成为男性不育诊断和治疗中的重要工具,推动辅助生殖领域的发展进程。更多还原

【Abstract】 Male infertility is a significant contributor to overall fertility issues,and sperm quality is crucial for successful fertilization and embryo development.Current methods of assessing sperm quality primarily rely on individual parameters such as motility,morphology,DNA integrity,and acrosomal reaction capability.However,these existing detection methods are often subjective,prone to errors,and fail to provide a comprehensive and accurate evaluation of sperm quality.To address these limitations,this study introduces an innovative multi-parameter detection method that integrates optical tweezers with deep learning technology,enabling comprehensive,non-invasive,and label-free analysis of individual sperm cells.The proposed method involves precise capture of individual sperm using optical tweezers,combined with deep learning models for a multidimensional assessment of sperm motility,volume,surface area,and DNA fragmentation.Initially,optical tweezers are used to capture individual sperm cells and analyze their longitudinal rolling frequency and the proportion of pausing duration,offering a more dynamic evaluation of sperm motility compared to conventional methods.Optical tweezers enable multiple assessments of the same sperm at different times,enhancing the accuracy of motility measurements.In addition,the study employs the U-Net network within deep learning frameworks to segment and perform three-dimensional surface reconstruction of the sperm head,both in free movement and under optical trapping.This approach provides precise measurements of sperm head volume and surface area,overcoming the limitations of ellipsoidal or spherical assumptions in conventional methods and delivering a more accurate depiction of the sperm’s true three-dimensional morphology.Furthermore,a virtual staining method for assessing sperm DNA fragmentation through generative adversarial networks(GANs)is developed in this study.This method eliminates the need for fluorescent dyes,which can affect sperm viability,and allows for label-free detection of DNA fragmentation levels by generating virtual stained images from brightfield images.This virtual staining technique offers a novel solution for evaluating sperm DNA integrity without the drawbacks associated with traditional staining methods.By integrating optical tweezers with deep learning technology,this study achieves multi-parameter detection of individual sperm cells under brightfield illumination.This robust technique provides comprehensive support for the assessment of sperm quality and holds promise for applications in assisted reproductive technologies,such as intracytoplasmic sperm injection(ICSI),where it can enhance the selection of high-quality sperm,thereby improving subsequent embryo development success rates.Additionally,it offers a new data-driven diagnostic tool for male infertility.Looking ahead,further optimization and refinement of optical tweezers and deep learning technologies could establish this method as an essential instrument in the diagnosis and treatment of male infertility,driving advancements in the field of assisted reproduction.更多还原