Robust Intensity Standardization in Brain Magnetic Resonance Images

Area of Science:

• Medical Imaging
• Neuroimaging
• Biomedical Engineering

Background:

• Magnetic Resonance Imaging (MRI) intensities lack fixed tissue-specific values, complicating automated image analysis.
• Variability in MRI acquisition parameters and time affects image intensity consistency.
• This inconsistency compromises the accuracy and efficiency of postprocessing tasks like segmentation and classification.

Purpose of the Study:

• To introduce a novel tissue-based standardization technique (SBST) for magnetic resonance (MR) brain images.
• To address the challenge of variable MRI intensity values across different scans and subjects.
• To enhance the reliability and accuracy of automated MR image analysis algorithms.

Main Methods:

• Developed a tissue-based standardization technique (SBST) utilizing histogram and tissue-specific intensity information.
• Computed three separate standardizing transformations for the three main brain tissues.
• Applied spline smoothing to create a continuous intensity mapping and assessed robustness using Dice index (>0.9) for segmentation overlap.

Main Results:

• SBST demonstrated robustness, maintaining high overlap (>0.9 Dice index) in automatic segmentation across diverse image sources.
• The technique improved inter-tissue discrimination and normalized tissue gray-level distributions towards Gaussianity.
• Quantitative comparisons showed favorable performance against existing literature approaches.

Conclusions:

• SBST effectively standardizes MR brain image intensities on a tissue-specific basis, enhancing consistency.
• The technique improves the accuracy and efficiency of automated image analysis tasks, particularly segmentation and classification.
• SBST offers a robust solution for overcoming intensity variability in MR neuroimaging across different acquisition settings.