To improve breast lesion detection in ultrasound images by addressing challenges such as speckle noise, acoustic artifacts, low contrast, and blurred lesion boundaries through a novel detection framework.
Approach:
Framework Development: A lightweight context-structure synergistic framework based on YOLOv13 was proposed, incorporating a Dual-Stream Mamba Aggregation (DSMA) module for contextual feature aggregation and a Structure-aware Axial Attention (SAA) module for modeling structural dependencies. The integration of these modules enhances feature representation while maintaining computational efficiency.
Key Findings:
The proposed method achieved competitive detection performance on the BUV and WH-BUS datasets, with specific metrics indicating its effectiveness.
The framework maintained 2.50M parameters, 6.4 GFLOPs, and 161.29 FPS, demonstrating its efficiency.
Ablation and cross-dataset analyses showed that DSMA and SAA provide complementary benefits for improved feature representation.
Interpretation:
The integration of contextual and structural features, as demonstrated by the results, enhances the detection capabilities of breast ultrasound imaging.
Limitations:
The study does not address the potential impact of varying image quality on detection performance, which could affect the generalizability of the results.
Further validation on larger and more diverse datasets may be necessary to confirm the findings.
Conclusion:
The proposed method offers a lightweight detection framework for breast ultrasound images by effectively modeling contextual and structural features.
