A Novel Framework for Joint Segmentation-Registration-Bias Correction in Low-Contrast Images

Abstract. Low-contrast medical images pose significant challenges for segmentation due to weak intensity gradients and ambiguous boundaries (e.g., in prostate or brain tumor regions). Traditional methods, such as thresholding or edge detection, often fail to simultaneously address joint registration, segmentation, and bias correction, leading to misalignment and inaccurate tissue delineation of low-contrast medical images. To address this challenge, we propose a novel variational model with synergistic mechanism that handles segmentation, registration, and bias correction simultaneously, with its innovation lying in utilizing image segmentation methods from simple scenarios to assist segmentation of low-contrast and other complex scenario images through registration and bias field correction. The theoretical analysis of the proposed model includes the existence of the solution and the convergence of the alternating minimization method (AMM). Numerical experiments validate that the model’s innovative unified framework and logarithmic transformation design significantly improve the accuracy of segmentation and registration for low-contrast images compared with state-of-the-art methods. The code of the algorithm in this paper is publicly available at https://github.com/zzyhp/S-R-B-SIIMS.git .