Automated segmentation of white matter fiber bundles using diffusion tensor imaging data and a new density based...

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Accurate segmentation of brain white matter (WM) fiber bundles is crucial for diagnosing neuropsychiatric disorders.
• Supervised methods are limited by the high cost of generating gold-standard data.
• Existing unsupervised methods often require pre-specified cluster numbers, which is impractical for complex datasets.

Purpose of the Study:

• To develop an unsupervised segmentation algorithm for brain white matter fiber bundles.
• To automatically segment fiber bundles using intrinsic diffusion tensor imaging (DTI) data.
• To avoid prior information or assumptions about data distributions.

Main Methods:

• Proposed a novel density-based clustering algorithm: Neighborhood Distance Entropy Consistency (NDEC).
• NDEC utilizes both local and global density information to discover natural clusters.
• Compared NDEC against state-of-the-art algorithms (chameleon, spectral clustering, DBSCAN, k-means) using JHU DTI data.

Main Results:

• NDEC achieved the highest average Dice ratio (0.94) and DBCV score (0.71) among all tested algorithms.
• Evaluated performance using Dice ratio (external) and DBCV index (internal) metrics.
• NDEC demonstrated superior performance in segmenting WM fiber bundles.

Conclusions:

• NDEC effectively segments clusters with arbitrary shapes and densities, suitable for WM fiber bundles.
• The algorithm is valuable for segmenting brain structures with indistinct boundaries.
• NDEC shows potential as a tool for broader pattern recognition and medical diagnostics.