Scholarlink

<p>In Vitro Fertilization (IVF) is a manual fertilization procedure in which embryos are cultured outside the human body, under controlled laboratory conditions. Conventionally, embryo transfer into a woman’s uterus is performed at the blastocyst stage, reached 5–6 days after fertilization. Many studies have shown a strong association between blastocyst morphological characteristics and clinical pregnancy rates. Therefore, the assessment of human blastocyst quality is a critical factor for the optimization and success of IVF procedures. Traditional blastocyst grading methods rely on human visual morphological assessment which is subjective and depends on the experience and knowledge of the clinical embryologist. To tackle this problem, automated embryo quality assessment has been proposed through accurate segmentation and identification of the blastocyst’s main components, namely the inner cell mass (ICM), trophectoderm (TE), and zona pellucida (ZP). This enables the automatic analysis of blastocyst morphological properties essential for selecting viable embryos. Recently, deep learning has become popular for medical image segmentation applications, leading to significant advances in the medical field, including IVF, by reducing human error and increasing decision-making accuracy. This paper deals with semantic segmentation of human blastocyst images using recent deep learning models, namely U-Net, DeepLabv3+, and DSE-Net, experimented with different backbones. The used CNNs, along with four vision transformer-based models, were compared on the only available blastocyst dataset. The results demonstrate the effectiveness of recent deep learning models for blastocyst segmentation, addressing challenges such as limited data availability, blastocyst size and shape variations, and noisy artifacts.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Semantic segmentation of human blastocyst images using deep CNNs and vision transformers

  • 期刊论文

  • 发表日期：

    2025年7月11日

查看全文
  • Wided S. Miled,
  • Rafik Ghali,
  • Sana Chtourou,
  • Moulay A. Akhloufi

摘要

In Vitro Fertilization (IVF) is a manual fertilization procedure in which embryos are cultured outside the human body, under controlled laboratory conditions. Conventionally, embryo transfer into a woman’s uterus is performed at the blastocyst stage, reached 5–6 days after fertilization. Many studies have shown a strong association between blastocyst morphological characteristics and clinical pregnancy rates. Therefore, the assessment of human blastocyst quality is a critical factor for the optimization and success of IVF procedures. Traditional blastocyst grading methods rely on human visual morphological assessment which is subjective and depends on the experience and knowledge of the clinical embryologist. To tackle this problem, automated embryo quality assessment has been proposed through accurate segmentation and identification of the blastocyst’s main components, namely the inner cell mass (ICM), trophectoderm (TE), and zona pellucida (ZP). This enables the automatic analysis of blastocyst morphological properties essential for selecting viable embryos. Recently, deep learning has become popular for medical image segmentation applications, leading to significant advances in the medical field, including IVF, by reducing human error and increasing decision-making accuracy. This paper deals with semantic segmentation of human blastocyst images using recent deep learning models, namely U-Net, DeepLabv3+, and DSE-Net, experimented with different backbones. The used CNNs, along with four vision transformer-based models, were compared on the only available blastocyst dataset. The results demonstrate the effectiveness of recent deep learning models for blastocyst segmentation, addressing challenges such as limited data availability, blastocyst size and shape variations, and noisy artifacts.