使用物体检测模型自动检测卵巢卵泡
Maya Haj Hassan1, Eric Reiter1,2, Misbah Razzaq3
1INRAE, CNRS, Université de Tours, PRC, Nouzilly, 37380, France.
Scientific reports
|December 31, 2024
概括
深度学习模型在组织学图像中准确地计算卵巢卵泡和黄体,改进了手工方法. 这加快了对女性生殖和潜在的药理干预措施的研究.
科学领域:
- 生殖生物学 生殖生物学
- 计算病理学计算病理学
- 生物医学成像学 生物医学成像学
背景情况:
- 卵巢卵泡发育对于女性生殖至关重要,涉及复杂的卵泡发育.
- 精确量化后期的卵巢结构 (腹囊,黄体) 对研究和药物开发至关重要.
- 在组织学中手动计数这些结构是耗时且容易出现错误的.
研究的目的:
- 评估深度学习模型的有效性,用于自动计数阴毛囊和红色体.
- 为此任务比较不同深度学习架构 (YOLO,RetinaNet) 的性能.
- 展示AI在提高生殖生物学研究准确性和效率方面的潜力.
主要方法:
- 开发了两个单阶段物体检测模型:YOLO和RetinaNet,利用各种骨干架构.
- 实施转移学习,早期停止和数据增强,以提高模型的通用性.
- 采样策略和焦点损失的应用,以解决数据集中的类不平衡问题.
主要成果:
- 在测试数据集上,RetinaNet的平均精度为83%,而YOLO的平均精度为75%.
- 这些模型在1000张图像的数据集上进行了训练和验证.
- 与手动计数相比,深度学习模型在速度和准确性方面取得了显著的改进.
结论:
- 深度学习提供了一种强大的工具,用于自动化对卵巢卵泡结构的量化.
- 这些人工智能驱动的方法可以加速生殖生物学和药物发现的研究.
- 开发的模型为传统的手动计数技术提供了更准确,更有效的替代方案.
相关概念视频
01:20Folliculogenesis
593
Folliculogenesis is the development of ovarian follicles, the specialized structures within the ovarian cortex where oogenesis, or egg development, occurs. This process is essential for female reproductive health and begins during fetal development when primordial follicles are formed. Each primordial follicle comprises a primary oocyte in the center, surrounded by a single layer of squamous pre-granulosa cells. These follicles remain dormant in late prophase I of meiosis until triggered by...
593
01:27Ovarian Cycle
993
The menstrual cycle includes a critical component known as the ovarian cycle, which undergoes two main phases each month—the follicular phase and the luteal phase. The follicular phase is variable and averaging around 14 days. Ovulation, triggered by a surge in luteinizing hormone (LH), marks the transition between the two phases. The second phase, the luteal phase, is relatively consistent, lasting approximately 14 days, and is marked by the activity of the corpus luteum. While a cycle...
993