囊胚孵化狀態為體外受精療程中評估胚胎發育程度的重要形態特徵之一。臨床上，early hatching 與 late hatching 多由胚胎師根據顯微影像進行人工判讀，雖具有臨床可解釋性，但容易受到影像品質、觀察角度、胚胎師經驗與分類標準差異影響。因此，如何將囊胚孵化狀態由主觀類別判讀轉換為客觀、可重現且可計算之量化指標，為本研究之主要問題。
本研究提出一套基於影像分割之囊胚孵化程度量化流程。首先，使用 U-Net 雙通道分割模型對囊胚顯微影像進行分割，分別取得透明帶內緣遮罩與胚胎外緣遮罩。接著，針對透明帶邊界可能出現之局部尖刺、凹陷或不連續問題，本研究提出中心點徑向局部中位數邊界修正方法。此方法以透明帶遮罩之中心點為基準，將邊界表示為不同角度下之半徑序列，並利用鄰近角度之局部中位數修正異常半徑，以提升透明帶邊界穩定性，同時避免橢圓擬合造成過度平滑。完成邊界修正後，本研究將胚胎外緣遮罩中位於修正後透明帶外側之區域定義為孵化區域，並進一步計算 area-based hatching ratio 與 diameter-based hatching ratio，作為囊胚孵化程度之幾何量化指標。
實驗結果顯示，中心點徑向局部中位數邊界修正方法在分割與邊界距離指標上優於 raw U-Net 與橢圓輕度修正方法。該方法可有效降低透明帶局部邊界誤差，並提升後續孵化比例計算之穩定性。在胚胎師一致性分析中，以 area-based hatching ratio 並採用 predefined cutoff = 1.0 進行 early/late hatching 分組時，模型分組與胚胎師人工判讀之整體一致率為 86.67%，顯示本研究所提出之量化指標能在多數案例中反映人工判讀邏輯。
此外，本研究進一步分析 hatching group 與臨床結果之關聯。結果顯示，area-based 與 diameter-based hatching group 皆與 PGT-A euploid status 呈現顯著統計關聯，其中 area-based hatching group 之 odds ratio 為 2.08，95% confidence interval 為 1.48–2.94，p-value 為 0.00003。相較之下，hatching group 與 implantation outcome 之關聯未達統計顯著，顯示囊胚孵化形態可能與胚胎染色體狀態具有一定關聯，但其與著床結果之關係仍可能受到母體條件、子宮內膜狀態、植入流程與其他臨床因子影響。
綜合而言，本研究建立了一套由 U-Net 分割、徑向邊界修正、孵化區域擷取至 hatching ratio 計算之完整影像量化流程。此方法可將傳統 early/late hatching 人工判讀轉換為可重現之幾何指標，並提供後續胚胎形態量化分析與臨床關聯研究之基礎。
關鍵字：囊胚孵化、透明帶、U-Net、影像分割、徑向邊界修正、孵化比例、胚胎形態量化;Blastocyst hatching status is an important morphological feature for evaluating embryonic development in in vitro fertilization. In clinical practice， early hatching and late hatching are commonly assessed by embryologists through visual inspection of microscopic images. Although this approach is clinically interpretable， it may be affected by image quality， viewing angle， embryologist experience， and differences in classification criteria. Therefore， converting blastocyst hatching status from subjective categorical assessment into an objective， reproducible， and computable quantitative index is the main problem addressed in this study.
This study proposes a segmentation-based framework for blastocyst hatching quantification. First， a dual-channel U-Net segmentation model is used to segment blastocyst microscopic images and generate two masks: the zona pellucida inner mask and the embryo outer mask. Since the predicted zona pellucida boundary may contain local spikes， concavities， or discontinuities， this study further proposes a centroid-based radial local median boundary refinement method. The proposed method represents the zona pellucida boundary as a sequence of radial distances from the centroid at different angles and corrects abnormal radii using the local median of neighboring angles. This design improves the stability of the zona pellucida boundary while avoiding the excessive smoothing that may occur with ellipse-based fitting. After boundary refinement， the hatched portion is defined as the region of the embryo outer mask located outside the corrected zona pellucida mask. The area-based hatching ratio and diameter-based hatching ratio are then calculated as geometric quantitative indicators of blastocyst hatching degree.
Experimental results show that the proposed radial local median boundary refinement method outperforms raw U-Net and ellipse-based correction in both segmentation overlap metrics and boundary distance metrics. The method effectively reduces local zona pellucida boundary errors and improves the stability of subsequent hatching ratio calculation. In the embryologist agreement analysis， using the area-based hatching ratio with a predefined cutoff of 1.0 to classify embryos into early and late hatching groups achieved an overall agreement of 86.67% with embryologist assessment. This result indicates that the proposed quantitative indicator can reflect embryologist interpretation in most cases.
This study further analyzed the association between hatching group and clinical outcomes. Both area-based and diameter-based hatching groups were significantly associated with PGT-A euploid status. For the area-based hatching group， the odds ratio was 2.08， with a 95% confidence interval of 1.48–2.94 and a p-value of 0.00003. In contrast， the association between hatching group and implantation outcome was not statistically significant. These findings suggest that blastocyst hatching morphology may be associated with chromosomal status， whereas its relationship with implantation outcome may be influenced by maternal conditions， endometrial status， embryo transfer procedures， and other clinical factors.
In summary， this study establishes a complete image-based quantification pipeline consisting of U-Net segmentation， radial boundary refinement， hatched portion extraction， and hatching ratio calculation. The proposed framework converts conventional embryologist-based early/late hatching assessment into a reproducible geometric indicator and provides a quantitative basis for subsequent embryo morphology analysis and clinical association studies.
