A New Variational Method for Bias Correction and Its Applications to Rodent Brain Extraction

Huibin Chang, Weimin Huang, Chunlin Wu, Su Huang, Cuntai Guan, Sakthivel Sekar, Kishore Kumar Bhakoo, Yuping Duan

Introduction

In preclinical research, rodent MRI is a vital tool for investigating disease progression and drug efficacy. However, researchers consistently face two significant technical hurdles: intensity inhomogeneity (the bias field) and brain extraction (skull stripping). High-field MRIs used for rodents often suffer from smooth, unwanted variations in brightness caused by radio-frequency coil imperfections. Furthermore, segmenting the brain from surrounding tissues like eyes and skin is difficult because tools designed for human anatomy often fail to account for the unique shapes and contrasts of rodent brains.

Existing workflows typically treat these as sequential steps—fixing the bias first, then extracting the brain. However, errors in bias correction often propagate to the extraction phase. To address this, the authors introduce HoL0MS (High-order and $L_0$ regularized Mumford-Shah), a novel variational model that simultaneously corrects intensity inhomogeneity and extracts brain tissue.

Methodology

The HoL0MS model treats an observed MRI image as a combination of a true piecewise-constant image, a smooth bias field, and noise. The algorithm's workflow is designed to optimize these components through a multi-resolution approach.

The HoL0MS algorithm workflow, illustrating the iterative minimization process and subsequent morphological steps to produce a brain mask.

Technical Innovations

  • $L_0$ Gradient Regularization: Unlike standard Total Variation (TV) methods that can cause "staircasing" artifacts, $L_0$ regularization encourages the "true" image to have a minimal number of sharp edges. This naturally separates the brain as a distinct, flat-intensity region.
  • High-Order (Hessian) Regularization: To model the bias field, the authors utilize the Hessian operator (second-order partial derivatives). This ensures the bias field is mathematically smooth, accurately reflecting the physical reality of coil sensitivity.
  • Multi-resolution Solver: To handle large 3D volumes efficiently, the model solves a coarse version of the image first, using it as a starting point for finer resolutions. Subproblems are solved using Fast Fourier Transforms (FFT), enabling high-speed iterations.

Evaluation and Results

The model was validated across three distinct rodent datasets (4.7T, 17.6T, and 9.4T) and a small set of human 3T T2-weighted images to demonstrate generalizability.

Visual representation of (a) a smooth bias field and (b) a target rodent brain MRI slice, the two primary components separated by the HoL0MS model.

Key Performance Metrics

  • Bias Correction: HoL0MS achieved a lower Coefficient of Variation (CV) across 20 different rodent brain tissues compared to industry-standard tools like N3 and N4.
  • Extraction Accuracy: The method achieved a Volume Overlap (VO) of approximately 92-94%. It significantly outperformed popular tools like the Brain Extraction Tool (BET) and Brain Surface Extractor (BSE), which frequently over-segmented or failed at the anterior/posterior regions of the rodent brain.
  • Computational Efficiency: Despite its 3D variational complexity, the FFT-based solver allows the system to process a high-resolution volume in under four minutes on a standard laptop.

Discussion and Conclusion

The primary insight of this work is that joint optimization is key: solving for the bias field and segmentation simultaneously allows the two processes to inform each other. A better estimate of the brain boundary leads to a more accurate bias field, and vice versa. Furthermore, $L_0$ minimization proved superior to TV for segmentation tasks, yielding genuine piecewise-constant solutions that make simple thresholding highly effective.

While highly effective for T2-weighted images, the authors note that HoL0MS currently struggles with T1-weighted MR images, where the contrast between the skull and cerebrospinal fluid (CSF) is often too low for the $L_0$ gradient to distinguish the boundary effectively. Future work will focus on adapting the model for diverse imaging modalities and integrating it into automated pipelines for longitudinal studies.

Summary updated

Abstract

Brain extraction is an important preprocessing step for further analysis of brain MR images. Significant intensity inhomogeneity can be observed in rodent brain images due to the high-field MRI technique. Unlike most existing brain extraction methods that require bias corrected MRI, we present a high-order and L <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> regularized variational model for bias correction and brain extraction. The model is composed of a data fitting term, a piecewise constant regularization and a smooth regularization, which is constructed on a 3-D formulation for medical images with anisotropic voxel sizes. We propose an efficient multi-resolution algorithm for fast computation. At each resolution layer, we solve an alternating direction scheme, all subproblems of which have the closed-form solutions. The method is tested on three T2 weighted acquisition configurations comprising a total of 50 rodent brain volumes, which are with the acquisition field strengths of 4.7 Tesla, 9.4 Tesla and 17.6 Tesla, respectively. On one hand, we compare the results of bias correction with N3 and N4 in terms of the coefficient of variations on 20 different tissues of rodent brain. On the other hand, the results of brain extraction are compared against manually segmented gold standards, BET, BSE and 3-D PCNN based on a number of metrics. With the high accuracy and efficiency, our proposed method can facilitate automatic processing of large-scale brain studies.
BibTeX
@article{chang2017new,
  author = {Huibin Chang and Weimin Huang and Chunlin Wu and Su Huang and Cuntai Guan and Sakthivel Sekar and Kishore Kumar Bhakoo and Yuping Duan},
  title = {{A New Variational Method for Bias Correction and Its Applications to Rodent Brain Extraction}},
  journal = {IEEE Trans. Medical Imaging},
  year = {2017},
  doi = {10.1109/tmi.2016.2636026},
}
Figure