Subject-specific abnormal region detection in traumatic brain injury using sparse model selection on high dimensional...

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Diffusion Tensor Imaging (DTI) provides insights into white matter microstructure.
• Mild Traumatic Brain Injury (mTBI) often presents subtle imaging abnormalities.
• Accurate detection of mTBI requires sophisticated analysis of complex neuroimaging data.

Purpose of the Study:

• To develop a robust method for identifying brain abnormalities in mTBI patients.
• To leverage multivariate Gaussian distribution modeling of diffusion tensor features.
• To enhance classification performance by incorporating brain region interactions.

Main Methods:

• Estimated a multivariate Gaussian distribution of diffusion tensor features from healthy subjects.
• Incorporated a neighborhood graph and L1 sparsity constraint on the precision matrix for modeling brain region interactions.
• Utilized the graphical LASSO algorithm for model estimation and Mahalanobis distance for classification.

Main Results:

• The proposed model with an apriori neighborhood graph significantly improved classification performance compared to models without interaction information or with fully connected priors.
• The Mahalanobis distance effectively discriminated between healthy and mTBI subjects.
• A method was developed to pinpoint specific brain regions contributing most to DTI profile abnormalities.

Conclusions:

• Integrating brain region interactions via a neighborhood graph enhances the sensitivity of DTI analysis for mTBI detection.
• The developed multivariate Gaussian model offers a powerful tool for identifying neuroimaging biomarkers of mTBI.
• The method facilitates the localization of brain regions affected by mTBI, aiding in clinical assessment.