Integrating Kolmogorov–Arnold Network and UNet Architecture for Reconstruction of 3-D Electrical Impedance Tomography to Assess the Efficacy of Tumor Treating Fields

This study presents an innovative KAN-UKAN framework that integrates the Kolmogorov-Arnold network (KAN) in the UNet architecture, aiming at achieving high-resolution 3-D electrical impedance tomography (EIT) image reconstruction for accurate assessment of the efficacy of tumor treating fields (TTFields). In this framework, KAN replaces the traditional fully connected layers in reconstruction networks, thereby improving the precision of mapping boundary voltages to conductivity distributions. Additionally, we strategically incorporate the KAN modules between the encoder and decoder of the UNet, referred to as UKAN, which strengthens the model’s ability to capture and learn complex nonlinear relationships in EIT data. Finally, the model’s overall performance and robustness were enhanced by combining the loss functions of the initial KAN predictions and the UKAN predictions with the ground truth. Simulation results on the lung cancer dataset and phantom experimental results show that the proposed KAN-UKAN framework significantly outperforms the four state-of-the-art reconstruction models. This superiority is evident in both quantitative metrics and qualitative analysis, demonstrating the accuracy and robustness of the framework in 3-D EIT image reconstruction. The methodology proposed in this study provides a novel and accurate way to assess the efficacy of TTFields and provides a new paradigm for clinical evaluation.