DTI Preprocessing Pipeline for Multimodal Data Fusion Analysis

This repository contains a comprehensive UNIX shell script for automated Diffusion Tensor Imaging (DTI) preprocessing using FSL (FMRIB Software Library) tools. The pipeline is specifically designed to prepare high-quality Fractional Anisotropy (FA) images for multimodal data fusion and unsupervised learning analysis using the Fusion ICA Toolbox (FIT).

Purpose

This preprocessing pipeline generates standardized, analysis-ready FA images optimized for advanced multimodal neuroimaging analyses including:

• tIVA (transposed Independent Vector Analysis)
• mCCA+jICA (multiset Canonical Correlation Analysis + joint Independent Component Analysis)
• Parallel ICA (PICA)
• Other multimodal fusion techniques available in the FIT toolbox

Pipeline Overview

The automated preprocessing script performs the following key steps to ensure FA images meet the quality standards required for multimodal fusion analysis:

1. Data Preparation & Quality Control

• Extracts b0 volumes from main DWI and reverse phase-encoding acquisitions
• Prepares data for distortion correction protocols

2. Advanced Distortion Correction

• TOPUP: Corrects susceptibility-induced distortions using opposite phase-encoding directions
• Essential for maintaining spatial accuracy required in multimodal registration

3. Brain Extraction & Masking

• BET: Performs optimized skull stripping (f=0.25) for DTI data
• Creates precise brain masks critical for accurate tensor fitting

4. Motion & Artifact Correction

• EDDY: Comprehensive correction for:
  • Head motion between volumes
  • Eddy current distortions
  • Residual susceptibility artifacts
• Ensures data quality suitable for sensitive ICA decompositions

5. DTI Model Fitting

• DTIFIT: Generates high-quality FA maps from corrected diffusion data
• Produces the primary input for multimodal fusion analyses

6. Spatial Standardization for Group Analysis

• FLIRT + FNIRT: Two-stage registration to MNI152 standard space
• Ensures spatial correspondence across subjects for group-level fusion
• Critical for valid cross-subject component analysis in tIVA and Parallel ICA

7. Optimization for FIT Toolbox

• Spatial smoothing (3.4mm = 8FWHM): Improves signal-to-noise ratio for ICA decomposition
• Standardized output format: Compatible with FIT toolbox input requirements
• Quality-controlled FA maps: Ready for integration with other modalities (fMRI, sMRI, etc.)

Multimodal Integration Ready

The final FA images (*_FA_final.nii.gz) are specifically prepared for:

• Cross-modal correspondence: Spatially aligned for fusion with fMRI, structural MRI, or other modalities
• Group-level ICA: Normalized and smoothed for stable component decomposition in Parallel ICA
• tIVA analysis: Optimized data quality for transposed Independent Vector Analysis across multiple datasets
• FIT toolbox compatibility: Formatted and processed according to toolbox requirements
• Statistical robustness: High-quality preprocessing reduces noise that could confound multimodal analyses

Requirements

• FSL (FMRIB Software Library)
• FIT (Fusion ICA Toolbox) for downstream analysis
• Standard DTI acquisition with reverse phase-encoding for distortion correction