Infertility affects millions of couples globally, with In Vitro Fertilization (IVF) serving as a critical treatment option. The morphological characteristics of human blastocyst components, such as the Inner Cell Mass (ICM) and Trophectoderm (TE) cells, are highly correlated with the success rate of IVF. However, conventional manual assessment of these components is labor-intensive, subjective, and prone to variability. To address these limitations, we propose an advanced deep learning framework, BlastocystMask, designed to automatically segment the internal structures of blastocyst images and enhance objectivity of morphological assessment. BlastocystMask is a two-stage instance segmentation framework. In stage one, it combines Res2Net and Deformable Convolutional Networks (DCN) for robust feature extraction and uses an FPN enhanced with CARAFE for multi-scale feature fusion. In stage two, PointRend refines the segmentation masks by focusing on uncertain regions, effectively addressing cell adhesion and blurred boundaries. Trained on a public human blastocyst dataset with expert annotations, BlastocystMask achieves superior performance, with a Dice coefficient of 91.9% and Jaccard index of 85.0%, outperforming existing methods. Ablation studies confirm that each module contributes to performance gains. BlastocystMask accurately identifies ICM regions and individual TE cells along the blastocyst’s equatorial plane, while introducing quantitative morphological parameters (e.g., cell size, shape uniformity) to complement subjective embryologist by automating segmentation and providing objective morphological metrics.