TC-KANRecon: High-Quality and Accelerated MRI Reconstruction via Adaptive KAN Mechanisms and Intelligent Feature Scaling

TL;DR

TC-KANRecon addresses the challenge of long MRI acquisition times by leveraging a conditional guided diffusion framework that fuses fully sampled k-space information with advanced feature modulation. The method introduces MF-UKAN for robust denoising and detail preservation, the Tok-KAN module for flexible feature extraction, and a dynamic clipping strategy to balance diversity and fidelity during sampling, all guided by MC-Model conditioning. Across fastMRI and SKM-TEA knee datasets, it achieves state-of-the-art reconstruction quality, especially under high-noise and low-sampling conditions, and demonstrates strong generalization across acceleration factors. Ablation studies confirm the critical roles of MF-UKAN, Tok-KAN, and dynamic clipping in delivering robust, high-fidelity reconstructions. The work has practical implications for faster, more reliable MRI while maintaining diagnostic detail, and it provides code for reproducibility.

Abstract

Magnetic Resonance Imaging (MRI) has become essential in clinical diagnosis due to its high resolution and multiple contrast mechanisms. However, the relatively long acquisition time limits its broader application. To address this issue, this study presents an innovative conditional guided diffusion model, named as TC-KANRecon, which incorporates the Multi-Free U-KAN (MF-UKAN) module and a dynamic clipping strategy. TC-KANRecon model aims to accelerate the MRI reconstruction process through deep learning methods while maintaining the quality of the reconstructed images. The MF-UKAN module can effectively balance the tradeoff between image denoising and structure preservation. Specifically, it presents the multi-head attention mechanisms and scalar modulation factors, which significantly enhances the model's robustness and structure preservation capabilities in complex noise environments. Moreover, the dynamic clipping strategy in TC-KANRecon adjusts the cropping interval according to the sampling steps, thereby mitigating image detail loss typicalching the visual features of the images. Furthermore, the MC-Model incorporates full-sampling k-space information, realizing efficient fusion of conditional information, enhancing the model's ability to process complex data, and improving the realism and detail richness of reconstructed images. Experimental results demonstrate that the proposed method outperforms other MRI reconstruction methods in both qualitative and quantitative evaluations. Notably, TC-KANRecon method exhibits excellent reconstruction results when processing high-noise, low-sampling-rate MRI data. Our source code is available at https://github.com/lcbkmm/TC-KANRecon.

Figures (3)

• Figure 1: The overall architecture of the TC-KANRecon model. Our model primarily comprises three key components: a VAE module for encoding and decoding images to reduce computational power, a conditional encoder module, MC-Model, for processing conditional images to improve targeted image generation, and a noise prediction backbone, MF-UKAN, which integrates the KAN network and MF model to enhance noise prediction.
• Figure 2: Details of the dynamic cropping strategy. After completing model training, a dynamically adaptive, linearly decreasing cropping technique that adapts dynamically at each step of the sampling process. The adjustment is based on the current progress of each sampling step. The adjustment is based on the current progress of each sampling step. It is applied to the noise estimates generated by the MF-UKAN module, resulting in MR images with improved visual effects.
• Figure 3: Compared with advanced reconstruction models on the fastMRI and SKM-TEA datasets in terms of visual effects. Enlarged detail images reveal that our approach better preserves texture details, indicating a stronger ability to restore fine features.