Magnetic resonance imaging bias field estimation and tissue segmentation via convolutional neural networks and multiplicative and additive intrinsic components optimization

Objective Magnetic Resonance Imaging (MRI) brain tissue segmentation is essential for diagnosing and treating neurological diseases; however, intensity inhomogeneity poses a significant challenge, particularly in data-limited scenarios. This study aims to explore an algorithm designed to robustly address intensity inhomogeneity for brain MRI tissue segmentation under dataset constraints. Methods We propose a two-stage framework that leverages both data-driven and knowledge-driven approaches. Initially, a data-driven model is employed for skull stripping, where a lightweight module, named multi-view dilated convolution attention (MDCA), is integrated into skip connections. The MDCA module eliminates the effect of intensity inhomogeneity enriched in shallow features at multiple scales, thus avoiding negative impact on deeper abstract features. Furthermore, we introduce multiplicative and additive intrinsic components optimization (MAICO) algorithm, which decomposes MRI images into their real anatomical structures, multiplicative and additive bias fields, and zero-mean Gaussian noise, enabling precise anatomical segmentation. Experiments on MRBrainS13 and MRBrainS18 public datasets involve randomly introducing intensity inhomogeneity to generate training, validation, and testing sets with 60%, 20%, and 20% splits, respectively. Segmentation performance is measured using Dice coefficients and compared to methods such as MICO, FSL, and UNet. An ablation study further validates the efficacy of the MDCA module. Results Our approach improves MRI brain tissue segmentation accuracy, achieving a mean Dice coefficient of 0.7733 across tissue types. With MDCA and MAICO, it reaches 0.8163 for white matter, 0.7402 for gray matter, and 0.7634 for cerebrospinal fluid, outperforming other algorithms. Additionally, the integration of the MDCA module in skip connections yields a 5% average accuracy boost. Conclusion This research effectively combined knowledge-driven and data-driven techniques to enhance MRI brain segmentation stability and accuracy, demonstrating strong potential for clinical application in managing intensity inhomogeneity in data-constrained settings.