Automated blood artery segmentation is a crucial work in biomedical image processing, and it faces several obstacles due to intricate vascular systems, anatomical variances, and a lack of annotated datasets. This paper provides an extensive literature analysis on machine learning methods for medical image segmentation, with a focus on the most recent approaches for different organs. The objective is to create a solid foundational model for vascular segmentation in Hierarchical Phase-Contrast Tomography (HiP-CT), a new imaging modality with high-resolution 3D imaging capabilities. Let’s move on to the more general setting of medical image segmentation, where deep Convolutional Neural Networks (CNNs), namely the UNet architecture, have become widely used due to recent advances. Nevertheless, UNet difficulty in capturing long-range dependencies leads to restrictions. In response, a unique model that combines the best features of transformers and CNNs is put forth in this paper. An attention gate, channel attention, and spatial normalization are three ways that the three-level attention (TLA) module is incorporated into the architecture to improve feature representation. Deep supervision and revised skip connections improve model performance by tackling issues like low-contrast tissue settings and hazy borders. The suggested model shows steady progress on various datasets, such as ultrasound and CT scan pictures, indicating its potential as a cutting-edge instrument for medical image segmentation.