LGFFM: A Localized and Globalized Frequency Fusion Model for Ultrasound Image Segmentation.

Area of Science:

• Medical Imaging and Diagnostics
• Artificial Intelligence in Healthcare
• Biomedical Signal Processing

Background:

• Accurate ultrasound image segmentation is crucial for disease screening and diagnosis.
• Neural network methods show promise but struggle with ultrasound image artifacts like noise and low resolution.
• Existing segmentation models lack flexibility for diverse scenarios, including organ and lesion segmentation.

Purpose of the Study:

• To develop a novel model for robust ultrasound image segmentation across various scenarios.
• To address limitations of current methods in handling ultrasound image quality issues.
• To improve the flexibility and generalization performance of segmentation models.

Main Methods:

• Proposed a Localized and Globalized Frequency Fusion Model (LGFFM) for ultrasound image segmentation.
• Introduced a Parallel Bi-Encoder (PBE) with Local Feature Blocks (LFB) and Global Feature Blocks (GLB) for enhanced feature extraction.
• Developed a Frequency Domain Mapping Module (FDMM) to capture texture and edge details, and a Multi-Domain Fusion (MDF) for integrating features.

Main Results:

• LGFFM demonstrated superior segmentation accuracy compared to state-of-the-art methods.
• The model showed improved generalization performance across diverse ultrasound datasets.
• Experiments were conducted on eight public ultrasound datasets covering four different types.

Conclusions:

• The proposed LGFFM effectively addresses challenges in ultrasound image segmentation.
• LGFFM offers enhanced accuracy and generalization, outperforming existing methods.
• This model holds significant potential for improving clinical applications of ultrasound imaging.