A systematic review of transformer-enhanced UNet architectures for 3D medical image segmentation: Trends, challenges, and the ATD-TᵣEE v framework.

In 30 seconds

This systematic review evaluates transformer-enhanced UNet architectures for 3D medical image segmentation, focusing on their application in tumor detection. Analyzing 87 articles from 2016-2025, the review found that hybrid models like TransUNet and FET-UNet achieved Dice scores between 0.81 and 0.96, indicating improved segmentation performance compared to traditional methods.

Key findings

• Hybrid models like TransUNet and FET-UNet achieved Dice scores of 0.81-0.96 in standard datasets.
• The review identified challenges such as GPU memory usage and data imbalance in current models.
• The ATD-TᵣEEv taxonomy was developed to analyze architectural trends and clinical applicability.
• Statistical validation remains lacking in many studies reviewed.

Why it matters

Accurate 3D medical image segmentation is crucial for effective tumor diagnosis and treatment. Understanding the advancements in transformer-enhanced models can help clinicians leverage these technologies for improved patient outcomes.