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

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Biomedical Engineering

Background:

• Rodent brain MRI analysis requires accurate preprocessing, including brain extraction.
• High-field MRI techniques often introduce significant intensity inhomogeneity (bias field).
• Existing methods typically require pre-corrected MRI data, limiting their applicability.

Purpose of the Study:

• To develop a unified model for simultaneous bias correction and brain extraction in rodent brain MRI.
• To address challenges posed by intensity inhomogeneity in high-field rodent MRI.
• To provide an efficient and accurate method for preprocessing rodent brain images.

Main Methods:

• A high-order and L0 regularized variational model was formulated in 3D for anisotropic voxel sizes.
• The model incorporates data fitting, piecewise constant, and smooth regularization terms.
• An efficient multi-resolution algorithm with closed-form solutions for subproblems was employed.

Main Results:

• The method was validated on 50 rodent brain volumes across 4.7T, 9.4T, and 17.6T MRI scans.
• Bias correction performance was compared against N3 and N4 methods using coefficient of variation on 20 tissues.
• Brain extraction accuracy was evaluated against manual segmentation and other algorithms (BET, BSE, 3D PCNN).

Conclusions:

• The proposed method effectively performs simultaneous bias correction and brain extraction.
• It demonstrates high accuracy and computational efficiency.
• This approach facilitates automated processing for large-scale rodent brain imaging studies.