Generalizable 2D Medical Image Segmentation Via Wavelet-Guided Spatial-Frequency Fusion Network | JoVE Visualize

Area of Science:

Medical Imaging
Computer Vision
Signal Processing

Background:

Medical image segmentation is crucial for diagnosis but struggles with domain shifts caused by varying imaging protocols and artifacts.
Existing methods often fail to integrate multi-scale spatial information with localized frequency details, hindering performance across different medical imaging domains.
This limitation necessitates novel approaches for robust medical image segmentation that can generalize across diverse datasets.

Purpose of the Study:

To propose a new framework, WGSF-Net, that unifies spatial and wavelet-frequency representations for improved medical image segmentation.
To address the challenge of domain generalization in medical image segmentation by effectively integrating multi-scale structural cues and localized frequency signatures.
To enhance the preservation of anatomical boundaries in medical images, particularly under domain shifts.

Main Methods:

Developed a novel framework employing wavelet-guided fusion to combine spatial and wavelet-frequency representations.
Introduced a wavelet-guided multi-scale attention mechanism to decompose features into directional subbands, capturing domain-invariant structural patterns.
Implemented an adaptive lateral fusion strategy for dynamic alignment of frequency-refined decoder features with spatially enhanced skip connections.

Main Results:

The proposed method achieved state-of-the-art performance on dermoscopy, ultrasound, and microscopy datasets, demonstrating strong generalization to unseen domains.
WGSF-Net showed significant improvements in Dice scores compared to previous methods, with gains up to 1.5% on dermoscopy, 2.0% on ultrasound, and 13.9% on microscopy in unseen settings.
The wavelet-guided spatial-frequency fusion approach effectively enhanced generalization capabilities in 2D medical image segmentation.

Conclusions:

Wavelet-guided spatial-frequency fusion is an effective strategy for improving the generalization of 2D medical image segmentation models.
The proposed WGSF-Net framework successfully addresses the limitations of existing methods in handling cross-domain variations.
This approach offers a promising direction for developing more robust and reliable medical image segmentation tools across diverse clinical applications.