EIT to CT Cross-Modality Translation Using Diffusion Transformer

Computed Tomography (CT) plays a crucial role in medical imaging due to its superior spatial resolution and diagnostic capability, particularly in pulmonary applications. However, the associated radiation risk, high equipment cost, and limitations in accessibility present significant challenges. In contrast, Electrical Impedance Tomography (EIT) offers a radiation-free, cost-effective, and portable imaging alternative, making it attractive in bedside and resource-limited settings. In this work, we demonstrate the feasibility of generating high-resolution synthetic CT (sCT) images from raw EIT voltages, enhancing EIT's clinical application by effectively unifying the benefits of both modalities. We propose an EIT-to-CT translation approach that leverages a diffusion transformer conditioned on the voltage measurements to guide sCT image generation through the reverse diffusion process. The model was trained on a synthetic dataset and evaluated on both the simulated and experimental data, resulting in anatomically coherent and visually realistic sCT images. Quantitative assessment demonstrates the method's effectiveness on both pixel-level and distribution-based metrics. It achieved a Correlation Coefficient of 0.9282, Peak Signal-to-Noise ratio of 24.64 dB, Feature Similarity Index of 0.8321, and deep Image Structure and Texture Similarity of 0.1289, indicating a high level of structural consistency. Similarly, it attained strong alignment with the targeted CT distribution with a Fréchet Inception Distance of 33.36 and Kernel Inception Distance of 0.0110. The clinical relevance of the sCT images was further validated through a downstream lung cancer classification task, where models trained on real CT showed comparable performance on sCT, and a pilot reader study, where clinical experts rated over 90% of the generated images as anatomically correct and realistic.