Saffari Hassan, Fathi Davood, Palay Peyman, Gourabi Hamid, Fathi Rouhollah
Department of Electrical and Computer Engineering, Tarbiat Modares University (TMU), Tehran, Iran.
Department of Embryology, Reproductive Biomedicine Research Center, Royan Institute for Reproductive Biomedicine, ACECR, Tehran, Iran.
Sci Rep. 2025 Jul 22;15(1):26532. doi: 10.1038/s41598-025-11810-4.
Despite advancements in in vitro fertilization (IVF) over the past 30 years, its outcome effectiveness remains low (20-40%). This study introduces a microfluidic-based machine learning framework to improve predictive accuracy in oocyte quality assessment. Immature oocytes were recorded as they passed through a custom-designed microfluidic channel under controlled flow. Using image processing, two biomechanical features-Cortical Tension (CT) and Deformation Index (DI)-were extracted. Additionally, oocyte diameter and the critical flow rate (Q)-defined as the minimum flow rate necessary for an oocyte to pass through the channel-were included as predictive variables. A dataset of 54 oocytes was labeled based on maturation, fertilization, and cleavage outcomes. Supervised learning models (Random Forest, Decision Tree, K-Nearest Neighbors, eXtreme Gradient Boosting, Logistic Regression, Naive Bayes, Support Vector Machines, and Light Gradient Boosting Machine were evaluated. Random Forest achieved the best classification accuracy: 76.10% (K-Fold) and 75.93% (Leave-One-Out). For unsupervised learning, K-Means, DBSCAN, Agglomerative Clustering, and Gaussian Mixture Models were applied. Among them, Agglomerative Clustering yielded the best performance (Silhouette = 0.49, Davies-Bouldin = 0.73), showing meaningful grouping patterns among oocytes. These results demonstrate that integrating biomechanical profiling with machine learning can significantly improve the objectivity and accuracy of oocyte quality prediction. This approach holds promise for enhancing embryo selection strategies in Assisted Reproductive Technology (ART) and optimizing In Vitro Fertilization (IVF) outcomes.
尽管在过去30年里体外受精（IVF）技术取得了进步，但其结果有效性仍然很低（20%-40%）。本研究引入了一种基于微流控的机器学习框架，以提高卵母细胞质量评估的预测准确性。未成熟卵母细胞在受控流动下通过定制设计的微流控通道时被记录下来。利用图像处理技术，提取了两个生物力学特征——皮质张力（CT）和变形指数（DI）。此外，卵母细胞直径和临界流速（Q）——定义为卵母细胞通过通道所需的最小流速——被纳入预测变量。基于成熟、受精和分裂结果，对54个卵母细胞的数据集进行了标记。对监督学习模型（随机森林、决策树、K近邻、极端梯度提升、逻辑回归、朴素贝叶斯、支持向量机和轻梯度提升机）进行了评估。随机森林取得了最佳分类准确率：76.10%（K折交叉验证）和75.93%（留一法）。对于无监督学习，应用了K均值、DBSCAN、凝聚聚类和高斯混合模型。其中，凝聚聚类表现最佳（轮廓系数=0.49，戴维斯-布尔丁指数=0.73），显示出卵母细胞之间有意义的分组模式。这些结果表明，将生物力学分析与机器学习相结合可以显著提高卵母细胞质量预测的客观性和准确性。这种方法有望增强辅助生殖技术（ART）中的胚胎选择策略，并优化体外受精（IVF）结果。
