RefLSM: Linearized Structural-Prior Reflectance Model for Medical Image Segmentation and Bias-Field Correction

TL;DR

The paper tackles robust medical image segmentation under severe intensity inhomogeneity and noise by introducing RefLSM, a variational level-set framework that leverages Retinex-based reflectance decomposition. It combines a linearized structural prior and a relaxed binary level-set, solved efficiently via ADMM to jointly segment and perform bias-field correction. The approach yields higher Dice and Precision across brain MR, cardiac MR, and ultrasound datasets, with strong bias correction and denoising performance and fast convergence. This work offers a practical, robust tool for clinical imaging tasks where illumination and bias distort segmentation boundaries.

Abstract

Medical image segmentation remains challenging due to intensity inhomogeneity, noise, blurred boundaries, and irregular structures. Traditional level set methods, while effective in certain cases, often depend on approximate bias field estimations and therefore struggle under severe non-uniform imaging conditions. To address these limitations, we propose a novel variational Reflectance-based Level Set Model (RefLSM), which explicitly integrates Retinex-inspired reflectance decomposition into the segmentation framework. By decomposing the observed image into reflectance and bias field components, RefLSM directly segments the reflectance, which is invariant to illumination and preserves fine structural details. Building on this foundation, we introduce two key innovations for enhanced precision and robustness. First, a linear structural prior steers the smoothed reflectance gradients toward a data-driven reference, providing reliable geometric guidance in noisy or low-contrast scenes. Second, a relaxed binary level-set is embedded in RefLSM and enforced via convex relaxation and sign projection, yielding stable evolution and avoiding reinitialization-induced diffusion. The resulting variational problem is solved efficiently using an ADMM-based optimization scheme. Extensive experiments on multiple medical imaging datasets demonstrate that RefLSM achieves superior segmentation accuracy, robustness, and computational efficiency compared to state-of-the-art level set methods.

Figures (19)

• Figure 1: Segmentation results from the LIF model for 2 different brain tumor MR images.
• Figure 2: Segmentation results from the RefLSM for 2 different brain tumor MR images. Row 1 and Row 3 :Results from the RefLSM. Row 2 and Row 4 : Ground truth.
• Figure 3: Image intensity inhomogeneity comparison with the bias field correction of the LIC model and the RefLSM. (a) is a brain MR image and (d) is its histogram. (b) is the result from the LIC model for bias field correction and (e) is its histogram. (c) is the result from the RefLSM for bias field correction and (f) is its histogram.
• Figure 4: Segmentation results from evaluated models for brain tumor MR images. Row 1:Original images and initial contours. Row 2-4: Results from the RESLS model, ALF model, L1 model, and the RefLSM. Row 6: Ground truth.
• Figure 5: Segmentation results from evaluated models for right ventricle. Column 1-7: Results from the ALF model, LoGRSF model, ABC model, RESLS model, ICTM model, FeaACM model and the RefLSM. Column 8: Ground truth.