A multivariate hypothesis testing framework for tissue clustering and classification of DTI data.

Area of Science:

• Medical Imaging
• Biophysics
• Computational Biology

Background:

• Diffusion Tensor MRI (DTI) is crucial for analyzing tissue microstructure.
• Accurate tissue classification in DTI data remains a challenge.
• Unsupervised methods offer potential for objective analysis without prior labels.

Purpose of the Study:

• To propose and validate a novel unsupervised algorithm for tissue clustering and classification in DTI data.
• To leverage the homogeneity of diffusion tensor distributions within voxels for improved accuracy.
• To assess the algorithm's effectiveness on both synthetic phantoms and biological samples.

Main Methods:

• The algorithm utilizes voxel-wise diffusion tensor homogeneity.
• An F-test, adapted from Hext and Snedecor, assesses tensor distribution similarity.
• Parsimonious model selection based on Schwarz Criterion assigns tissue types to diffusion models.

Main Results:

• The unsupervised clustering effectively identified distinct regions of interest (ROIs).
• Validation was performed using numerical phantoms and excised rat and pig spinal cords.
• The classification approach demonstrated robust performance in differentiating tissue types.

Conclusions:

• The proposed unsupervised algorithm offers an effective approach for DTI-based tissue clustering and classification.
• This method has potential applications in neuroimaging and spinal cord research.
• The reliance on diffusion tensor homogeneity provides a robust basis for tissue characterization.