.png){kind=icon}
.png){kind=icon}
Return
Wavelet spectral-aware Kolmogorov-Arnold Network for organ and tumor segmentation
.png){kind=icon}
PRE
AI
J
L
L
F
X
R
Z
Q
H
DOI：10.1016/j.compmedimag.2026.102784.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• We propose MedWKAN, a novel 3D medical segmentation model integrating Kolmogorov-Arnold Network with wavelet spectral-aware mechanism, achieving precise modeling of complex organ structures and tumors. • The proposed Gram Polynomial-inspired Wavelet Kolmogorov–Arnold (GWKA) module reconstructs the KAN operator through a polynomial nonlinear mapping on the low-frequency components, enabling dynamic global anatomical modeling while preserving high-frequency details, effectively alleviating nonlinear representation and detail loss in organ and tumor segmentation. • The Multi-Residual Collaborative Module (MRCM) leverages hierarchical convolution–normalization and progressive residual fusion to effectively mitigate the semantic gap between the encoder and decoder, enabling semantic alignment and stable feature transmission in skip connections. • MedWKAN achieves state-of-the-art performance across six public datasets, outperforming ten segmentation methods in Dice and IoU metrics, demonstrating both strong generalization and clinical applicability.
Journal
IF:
4.9
Papers:
2.4K
Citations:
5.0K
Organization
Cited Papers