Objective:

To propose a novel sequence-guided dual-branch segmentation network (TGMS-UNet) for accurate endometrial segmentation in ultrasound images, specifically addressing challenges such as noise interference and boundary ambiguity.

Approach:

• Sequence-Guided Framework: TGMS-UNet encodes endometrial contours into distance-angle-based sequences, which help to resolve ambiguities in ultrasound images by incorporating clinical reasoning.
• Feature Correction and Adaptive Fusion: A feature correction and adaptive fusion module is designed to mitigate multi-scale feature misalignment, allowing for dynamic adjustment of fusion ratios based on the input data.

Key Findings:

• TGMS-UNet outperforms state-of-the-art segmentation methods across multiple metrics.
• The model effectively incorporates clinical prior knowledge to enhance segmentation accuracy.

Interpretation:

TGMS-UNet demonstrates improved segmentation performance by integrating geometric contour information and addressing feature misalignment, as evidenced by the results.

Limitations:

• The study may be limited by the specific datasets used for training and validation, which may not represent the full diversity of clinical scenarios.
• Further validation on a wider range of clinical datasets is necessary to generalize the findings.

Conclusion:

TGMS-UNet offers a solution for endometrial segmentation in ultrasound imaging, aiming to enhance diagnostic objectivity.