Automatic brain extraction methods for T1 magnetic resonance images using region labeling and morphological...

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computer Vision

Background:

• Accurate brain extraction is crucial for neuroimaging analysis.
• Existing brain extraction methods (BEM) have limitations in accuracy and processing speed.
• Unsupervised and knowledge-based approaches are desirable for robust brain segmentation.

Purpose of the Study:

• To propose two novel, simple, unsupervised, and knowledge-based brain extraction methods (BEM).
• To evaluate the performance of the proposed BEM against established methods.
• To address limitations of previous methods, particularly in cases with complex brain structures.

Main Methods:

• Adaptive intensity thresholding on MRI scans to obtain a binary image.
• Anatomical labeling based on scalp and skull boundaries.
• Run-length scheme and morphological operations, incorporating a 3D approach to handle complex brain components.

Main Results:

• Proposed methods achieved high accuracy with an average Dice similarity index of 0.938 and specificity of 0.992.
• Outperformed FSL's BET and BrainSuite's BSE, with results comparable to Model-based Level Sets (MLS).
• Demonstrated faster processing time (≈1s/slice) and lower false positive rates (0.075) than existing methods, even on scans with tumors or lesions.

Conclusions:

• The proposed brain extraction methods offer a robust, efficient, and accurate alternative for neuroimaging analysis.
• These methods are effective even in challenging cases where other techniques fail.
• The unsupervised and knowledge-based nature makes them broadly applicable across different imaging orientations and datasets.