Generalized method for partial volume estimation and tissue segmentation in cerebral magnetic resonance images

Area of Science:

• Medical Imaging
• Neuroimaging
• Image Processing

Background:

• Partial volume averaging (PVA) is a significant artifact in magnetic resonance imaging (MRI) that hinders accurate segmentation of cerebral structures.
• Existing segmentation methods, such as Gaussian mixture models and multispectral techniques, have limitations with non-Gaussian noise, pathology, and modeling partial volume fractions directly.

Purpose of the Study:

• To develop a robust partial volume (PV) fraction estimation approach for cerebral MRI segmentation.
• To overcome the limitations of traditional methods by not relying on predefined intensity models or multispectral data.

Main Methods:

• A novel PV fraction estimation method is proposed, directly utilizing information from individual MRI scans.
• The approach employs an adaptively defined global edge map, derived from the relationship between edge content and PVA.
• The estimated PV fraction map is then used for high-accuracy segmentation of anatomy and pathology.

Main Results:

• The developed method accurately estimates the PV fraction in both simulated and real MRI datasets, including those with Gaussian and non-Gaussian noise.
• Segmentation results demonstrate subvoxel accuracy and robustness, even in the presence of pathology.
• The new approach shows significant advantages over traditional model-based segmentation techniques.

Conclusions:

• The proposed PV fraction estimation method offers a robust and accurate solution for segmenting cerebral MRI data.
• This technique effectively addresses the challenges posed by partial volume averaging, improving diagnostic capabilities.
• The findings highlight the potential of this method for enhanced neuroimaging analysis.