AdaKAN: A dual-branch adaptive Kolmogorov-Arnold network for medical image segmentation

2026-07-24

AI Summary

Medical image segmentation faces significant challenges due to the anatomical complexity and large modality variations, which hinder existing methods from simultaneously modeling long-range dependencies and fine-grained local features. To address this, this work proposes AdaKAN—a dual-branch adaptive Kolmogorov–Arnold Network that integrates convolutional layers with a novel EffiKAN module within a U-shaped architecture to jointly capture global contextual dependencies and local details. The core innovations include KAN layers based on Bernstein polynomial activations, a channel-wise adaptive scaling mechanism, and an efficient attention module, collectively overcoming key limitations of both CNNs and Transformers. Extensive experiments demonstrate that AdaKAN achieves state-of-the-art segmentation accuracy, significantly outperforming current baselines across multiple medical imaging datasets.

Abstract

Medical image segmentation is a fundamental task in computer-aided diagnosis, yet it remains challenging due to the complexity of anatomical structures and the variability across imaging modalities. In this paper, we propose AdaKAN, an Adaptive Kolmogorov-Arnold Network (KAN) that synergistically integrates convolutional operations with a novel efficient KAN (EffiKAN) block, comprised of an efficient attention mechanism and an adaptive KAN (AdaptKAN) module. This module features a dual-branch design: one branch employs a KAN layer with Bernstein polynomial activations for globally smooth and stable function approximation, while the other branch performs channel-wise refinement through projection operations and adaptive scaling. AdaKAN adopts a U-shaped architecture that effectively captures both long-range dependencies and fine-grained local features, overcoming the limitations of conventional convolutional and Transformer-based segmentation models. Skip connections are employed to preserve spatial details during encoding and facilitate accurate reconstruction during decoding. Extensive experiments conducted on diverse medical imaging datasets demonstrate that AdaKAN achieves state-of-the-art performance in segmentation accuracy.

Problem

• medical image segmentation
• anatomical complexity
• imaging modality variability
• computer-aided diagnosis

Innovation

• Kolmogorov-Arnold Network
• dual-branch architecture
• Bernstein polynomial activation
• efficient attention mechanism
• medical image segmentation