An ultrasonic image segmentation method based on frequency domain adaptive fusion and structure-aware self-distillation
By constructing a frequency-domain adaptive fusion encoder and a structure-aware self-prompting generator, the problems of insufficient generalization ability and low automation in ultrasound image analysis are solved, achieving efficient and automated segmentation of ultrasound images and improving the accuracy and consistency of segmentation results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DALIAN UNIV OF TECH
- Filing Date
- 2026-06-09
- Publication Date
- 2026-07-17
AI Technical Summary
Existing deep learning models lack generalization ability in ultrasound image analysis, making it difficult to adapt to varied anatomical structures. Furthermore, the automation and stability of the interactive segmentation process are insufficient, resulting in poor consistency of segmentation results and increasing the workload of physicians.
A frequency-domain adaptive fusion encoder and a structure-aware self-prompt generator are constructed. Multi-scale modeling is performed through wavelet decomposition, frequency-domain adaptation layer and multi-scale frequency-domain information storage module. Geometric structural features are extracted by the structure-aware branch module to generate a sparse self-prompt point set. The result is achieved by deep fusion through multi-domain fusion decoder to realize automated segmentation.
It significantly improves the model's adaptability and accuracy in different clinical scenarios, achieves full automation of the segmentation process, and improves the consistency and clinical usability of segmentation results.
Smart Images
Figure CN122415652A_ABST