Structure and Smoothness Constrained Dual Networks for MR Bias Field Correction

2025-07-01

AI Summary

MR image intensity inhomogeneity, caused by hardware limitations, severely impairs qualitative and quantitative medical analysis. Existing unsupervised deep learning methods model only global appearance, neglecting structural priors and smoothness constraints on the bias field, leading to geometric distortions and inaccurate correction. This paper proposes S2DNets, a self-supervised dual-network framework that— for the first time in unsupervised MR bias field correction—jointly incorporates piecewise structural constraints and bias field smoothness regularization. Through cooperative dual-branch modeling, S2DNets achieves feature disentanglement and error suppression. The method is trained end-to-end using a composite loss integrating structural similarity and smoothness regularization terms. Extensive experiments on both simulated and clinical MR datasets demonstrate that S2DNets significantly outperforms conventional and state-of-the-art deep learning methods, yielding markedly improved visual quality. Moreover, downstream segmentation tasks validate the practical utility and generalizability of its corrected images.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Computer Vision: Bias, Fairness & Privacy
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Social Networks and Social Media: Fairness and bias in social network and social media analysis
• Web Mining and Content Analysis: Bridging structured and unstructured data

Abstract

MR imaging techniques are of great benefit to disease diagnosis. However, due to the limitation of MR devices, significant intensity inhomogeneity often exists in imaging results, which impedes both qualitative and quantitative medical analysis. Recently, several unsupervised deep learning-based models have been proposed for MR image improvement. However, these models merely concentrate on global appearance learning, and neglect constraints from image structures and smoothness of bias field, leading to distorted corrected results. In this paper, novel structure and smoothness constrained dual networks, named S2DNets, are proposed aiming to self-supervised bias field correction. S2DNets introduce piece-wise structural constraints and smoothness of bias field for network training to effectively remove non-uniform intensity and retain much more structural details. Extensive experiments executed on both clinical and simulated MR datasets show that the proposed model outperforms other conventional and deep learning-based models. In addition to comparison on visual metrics, downstream MR image segmentation tasks are also used to evaluate the impact of the proposed model. The source code is available at: https://github.com/LeongDong/S2DNets}{https://github.com/LeongDong/S2DNets.

Problem

• Corrects MR image intensity inhomogeneity for better diagnosis
• Enhances structural details while ensuring bias field smoothness
• Improves accuracy in downstream MR image segmentation tasks

Innovation

• Dual networks with structure constraints
• Smoothness-constrained bias field correction
• Self-supervised MR image enhancement