GLFC: Unified Global-Local Feature and Contrast Learning with Mamba-Enhanced UNet for Synthetic CT Generation from CBCT

TL;DR

The paper addresses the challenge of generating high-quality synthetic CT images from CBCT by tackling global and local feature and contrast learning. It introduces GLFC, a framework combining a Mamba-Enhanced UNet (MEUNet) with a Multiple Contrast Loss (MCL). MEUNet integrates Visual State Space (VSS) Mamba blocks into skip connections of a high-resolution UNet to capture long-range context while preserving local detail, and MCL adds both a global loss over the full HU range and two local-window losses targeting soft tissue and bone contrasts, defined as $L_{mcl}=L_{glob}+L_{soft}+L_{bone}$. On SynthRad2023, GLFC achieves state-of-the-art performance, increasing SSIM from $77.91\%$ (baseline) to $91.50\%$ and delivering higher PSNR values, with MEUNet also offering a markedly smaller model size relative to baselines. The work has practical significance for radiotherapy planning and tumor assessment by providing more accurate, tissue-specific sCT images, and includes public code for reproducibility.

Abstract

Generating synthetic Computed Tomography (CT) images from Cone Beam Computed Tomography (CBCT) is desirable for improving the image quality of CBCT. Existing synthetic CT (sCT) generation methods using Convolutional Neural Networks (CNN) and Transformers often face difficulties in effectively capturing both global and local features and contrasts for high-quality sCT generation. In this work, we propose a Global-Local Feature and Contrast learning (GLFC) framework for sCT generation. First, a Mamba-Enhanced UNet (MEUNet) is introduced by integrating Mamba blocks into the skip connections of a high-resolution UNet for effective global and local feature learning. Second, we propose a Multiple Contrast Loss (MCL) that calculates synthetic loss at different intensity windows to improve quality for both soft tissues and bone regions. Experiments on the SynthRAD2023 dataset demonstrate that GLFC improved the SSIM of sCT from 77.91% to 91.50% compared with the original CBCT, and significantly outperformed several existing methods for sCT generation. The code is available at https://github.com/HiLab-git/GLFC

Figures (2)

• Figure 1: Overview of our Global-Local Feature and Contrast (GLFC) learning framework for synthetic CT generation.
• Figure 2: Visual comparison of sCT obtained by different methods. Images in the three rows are visualized with a global intensity window, soft tissue window and bone window, respectively. Local differences are highlighted by circles.