Accuracy and reliability of diffusion imaging models.

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Computational Neuroscience

Background:

• Diffusion imaging is crucial for non-invasively assessing brain white matter structure and integrity.
• Current diffusion imaging methods can struggle with accuracy and reliability due to data quantity and analysis choices.
• Overfitting, estimating non-existent fibers, is a known issue with existing crossing-fiber models.

Purpose of the Study:

• To evaluate the accuracy and reliability of common diffusion imaging methods.
• To investigate the impact of data quantity and analysis techniques on diffusion imaging outcomes.
• To develop and validate a novel method to mitigate overfitting in diffusion imaging analysis.

Main Methods:

• Simulations and highly sampled individual-specific diffusion data (927-1442 diffusion weighted images) were used for evaluation.
• Common crossing-fiber methods (FSL's BedpostX, DSI Studio's CSA-QBI, MRtrix3's CSD) were assessed.
• A new Bayesian Multi-tensor Model-selection (BaMM) method was developed and applied to FSL's BedpostX.

Main Results:

• Existing crossing-fiber methods showed overfitting (excess fiber estimation) with insufficient data or model mismatches.
• The novel BaMM method demonstrated robustness against overfitting.
• BaMM exhibited high reliability and superior crossing-fiber accuracy as diffusion data increased.

Conclusions:

• Sufficient diffusion imaging data and overfitting-resistant analysis methods are essential for enhancing precision.
• The BaMM method offers improved accuracy and reliability for white matter tractography.
• Diffusion imaging requires less data for reliability in clinical applications like surgical planning compared to functional MRI.