RHLS: A Robust Hybrid Level Set Model Using Global-Local Signed Energy-Based Pressure Force for Medical Image Segmentation

Medical image segmentation often encounters significant challenges due to noise and intensity inhomogeneity. While Level Set Models (LSMs) are widely used for segmentation, their effectiveness in these scenarios remains limited. This paper proposes a novel Robust Hybrid Level Set (RHLS) model specifically designed to address these limitations in medical image segmentation. The RHLS model tackles noise and intensity inhomogeneity by incorporating both global and local information into its energy function through a Signed Pressure Force (SPF) formulation. It leverages three distinct energy terms: Global Region-Based Term (GRT) which enhances robustness against various noise types using the Huber loss function, Global-Local Region-Based Term (GLRT) which promotes accurate contour propagation while mitigating noise influence, and Local Region-Based Term (LRT) which improves detection of faint edges and mitigates intensity inhomogeneity. These terms, combined with a regularization term, enable the RHLS model to exploit the strengths of both region-based and edge-based approaches. This leads to superior handling of noise and intensity variations compared to existing methods. Extensive evaluation using synthetic and real medical images demonstrates the effectiveness of the RHLS model. It achieves an average accuracy of 97.2% and 95.6% on synthetic and real medical data, respectively, using the Dice similarity measure. These results validate the model’s ability to handle various noise types and intensity variations, showcasing its potential for improved medical image segmentation tasks.