3D+t brain MRI segmentation using robust 4D Hidden Markov Chain.

Area of Science:

• Medical Imaging
• Neuroscience
• Computer Vision

Background:

• Diagnosing brain disorders often relies on complex imaging analysis.
• Automatic methods for brain disorder diagnosis are evolving but face challenges.
• Longitudinal Magnetic Resonance Imaging (MRI) provides valuable data for tracking brain changes over time.

Purpose of the Study:

• To develop and validate an automated method for segmenting brain structures in longitudinal 3D multi-modal MRI.
• To improve the robustness of brain structure segmentation and detect Multiple Sclerosis (MS) lesions as outliers.
• To leverage temporal and spatial information for enhanced diagnostic capabilities.

Main Methods:

• A novel bias field correction technique using an adaptation of the Hidden Markov Chain (HMC) was employed.
• The Hidden Markov Chain (HMC) incorporated temporal correlation alongside spatial neighborhood information.
• The Trimmed Likelihood Estimator (TLE) was utilized to enhance segmentation robustness and identify Multiple Sclerosis (MS) lesions.
• The method was validated on 3D+t (three-dimensional plus time) brain MRI datasets.

Main Results:

• The proposed method demonstrated effective segmentation of principal brain structures in longitudinal MRI.
• The integration of temporal correlation within the HMC improved bias field correction.
• The Trimmed Likelihood Estimator (TLE) successfully aided in detecting Multiple Sclerosis (MS) lesions as outliers.
• Validation on 3D+t brain MRI confirmed the method's performance.

Conclusions:

• The developed method offers a robust approach for brain structure segmentation in longitudinal MRI.
• This technique has the potential to improve the diagnosis and monitoring of brain disorders, including Multiple Sclerosis (MS).
• The combined use of HMC with temporal correlation and TLE represents a significant advancement in automated neuroimaging analysis.