Semi-KAN: KAN Provides an Effective Representation for Semi-Supervised Learning in Medical Image Segmentation

TL;DR

This work tackles annotation scarcity in medical image segmentation by introducing Semi-KAN, a semi-supervised framework that integrates Kolmogorov-Arnold Networks (KANs) into a U-Net backbone to enhance high-level semantic representations with fewer parameters. The method combines a shared encoder with multi-decoder branches and KAN-based blocks at the encoder bottleneck and decoder tops, paired with an uncertainty-estimation-based consistency loss to leverage unlabeled data. Across four public datasets, Semi-KAN outperforms state-of-the-art SSL methods at low labeling rates and approaches fully supervised performance, while maintaining lower computational cost. Additionally, the approach demonstrates interpretability through semantic feature alignment with anatomical boundaries and characteristic B-spline activation patterns, highlighting the practical potential of KANs in medical image analysis.

Abstract

Deep learning-based medical image segmentation has shown remarkable success; however, it typically requires extensive pixel-level annotations, which are both expensive and time-intensive. Semi-supervised medical image segmentation (SSMIS) offers a viable alternative, driven by advancements in CNNs and ViTs. However, these networks often rely on single fixed activation functions and linear modeling patterns, limiting their ability to effectively learn robust representations. Given the limited availability of labeled date, achieving robust representation learning becomes crucial. Inspired by Kolmogorov-Arnold Networks (KANs), we propose Semi-KAN, which leverages the untapped potential of KANs to enhance backbone architectures for representation learning in SSMIS. Our findings indicate that: (1) compared to networks with fixed activation functions, KANs exhibit superior representation learning capabilities with fewer parameters, and (2) KANs excel in high-semantic feature spaces. Building on these insights, we integrate KANs into tokenized intermediate representations, applying them selectively at the encoder's bottleneck and the decoder's top layers within a U-Net pipeline to extract high-level semantic features. Although learnable activation functions improve feature expansion, they introduce significant computational overhead with only marginal performance gains. To mitigate this, we reduce the feature dimensions and employ horizontal scaling to capture multiple pattern representations. Furthermore, we design a multi-branch U-Net architecture with uncertainty estimation to effectively learn diverse pattern representations. Extensive experiments on four public datasets demonstrate that Semi-KAN surpasses baseline networks, utilizing fewer KAN layers and lower computational cost, thereby underscoring the potential of KANs as a promising approach for SSMIS.

Figures (7)

• Figure 1: Performance and efficiency comparison of U-Net, Att-Unet, U-KAN, and Semi-KAN on the GLAS dataset. GFLOPs are measured during model inference with an input size of 256 $\times$ 256.
• Figure 2: Workflow of the proposed Semi-KAN. Semi-KAN utilizes a shared encoder and a multi-decoder network architecture with KANs. The decoders apply different upsampling strategies as perturbations, and the average of the predicted probability maps from the outputs is used as pseudo-annotations (intermediary supervision annotations).
• Figure 3: The architecture of KAN-Conv Block.
• Figure 4: The segmentation results of Semi-KAN on the GlaS, BUSI and CVC datasets.
• Figure 5: Qualitative comparisons of Semi-KAN, MC-net+ and AC-MT on the ACDC datasets.