FAN-Unet: Enhancing Unet with vision Fourier Analysis Block for Biomedical Image Segmentation
Jiashu Xu
TL;DR
FAN-Unet tackles the dual challenge of long-range dependency modeling and periodic pattern learning in biomedical image segmentation by integrating a Vision-FAN backbone with the U-Net architecture. The method combines position encoding, window-based self-attention, and the FANLayer2D to capture global context and periodic features, trained with a joint cross-entropy and Dice loss ($L_B=(1-\alpha)\cdot CE_B + \alpha\cdot Dice_B$, $\alpha=0.5$). Experimental results on ISIC2017 and ISIC2018 demonstrate competitive mIoU and Dice scores, with FAN-Unet outperforming several SOTA baselines and showing the effectiveness of the Vision-FAN Block and Fourier-based layers. This work highlights the practicality of marrying Fourier analysis with deep learning for robust medical image segmentation and suggests broad applicability to other imaging modalities.
Abstract
Medical image segmentation is a critical aspect of modern medical research and clinical practice. Despite the remarkable performance of Convolutional Neural Networks (CNNs) in this domain, they inherently struggle to capture long-range dependencies within images. Transformers, on the other hand, are naturally adept at modeling global context but often face challenges in capturing local features effectively. Therefore, we presents FAN-UNet, a novel architecture that combines the strengths of Fourier Analysis Network (FAN)-based vision backbones and the U-Net architecture, effectively addressing the challenges of long-range dependency and periodicity modeling in biomedical image segmentation tasks. The proposed Vision-FAN layer integrates the FAN layer and self-attention mechanisms, leveraging Fourier analysis to enable the model to effectively capture both long-range dependencies and periodic relationships. Extensive experiments on various medical imaging datasets demonstrate that FAN-UNet achieves a favorable balance between model complexity and performance, validating its effectiveness and practicality for medical image segmentation tasks.