LUCIDA: Low-Dose Universal-Tissue CT Image Domain Adaptation for Medical Segmentation

Yixin Chen, Xiangxi Meng, Yan Wang, Shuang Zeng, Xi Liu, Zhaoheng Xie

Abstract: Accurate segmentation in low-dose CT scans remains a challenge in medical imaging, primarily due to the high annotation costs. This study introduces LUCIDA, a Low-dose Universal-tissue CT Image Domain Adaptation model. LUCIDA operates under an unsupervised framework, eliminating the need for LDCT annotations. A novelty of LUCIDA lies in its integration of the Weighted Segmentation Reconstruction (WSR) module with a Fourier-based UNet (F-UNet), which not only establishes a linear relationship between prediction maps and ROI-based reconstructed images but also enhances segmentation accuracy through frequency domain adaptation of LDCT images. LUCIDA improves the accuracy of prediction maps, facilitating a new domain adaptation framework. Through extensive evaluation experiments, LUCIDA has demonstrated its effectiveness in accurately recognizing a wide array of tissues, significantly outperforming conventional methods. Additionally, we present the LUCIDA Ensemble model, which achieves performance comparable to supervised learning models in organ segmentation, capable of recognizing up to 112 tissue types.