🧭 Workshop Week 2 Outline

1. Diffusion MRI Protocol
   • Best practices and recommendations for diffusion MRI protocol
2. Preprocessing DWI Data
   • Motion correction & eddy current distortion correction
   • Susceptibility artifact correction
   • Other quality issues
3. Diffusion Models: Resolving Fibers and Measuring Anisotropy
   • Overview of DTI, GQI/QSDR

---

🧭 Session 1: DSI Studio Pipeline (10 minutes)

🖐️ Hands-On Practice

NIFTI/bids OpenNeuro ds002087: datasets with and without deliberate head movements for detection and imputation of dropout in diffusion MRI

1. Convert NIFTI to SRC
2. SRC to FIB

---

🧭 Session 2: Diffusion MRI Protocol (10 minutes)

✅ Protocol Checklist 1: Sufficient Diffusion Sensitivity

• Use b-value > 1,000 s/mm² for human studies
• ⚠️ No post-processing can recover low diffusion sensitivity

📌 Example: HCP-YA dataset
b-values: 0, 1000, 2000, 3000 s/mm²
b-vector: (0, 0, 1)

---

✅ Protocol Checklist 2: Isotropic Resolution

• Ensure slice thickness ≈ in-plane resolution
• If anisotropic, consider regridding to isotropic before analysis

📌 Example: OpenNeuro ds005849
In-plane: 1.75 mm
Slice thickness: 2.7 mm → ⚠️ Not isotropic

---

✅ Protocol Checklist 3: Reverse-Phase b0 Image

• Used to correct susceptibility distortions, especially in frontal and temporal lobes
• Some tools allow T1w-based correction, but b0 pairs are preferred
• Sequence-based corrections (e.g., readout segmentation) are also available

📌 Source: FSL Topup User Guide

Before Correction
(Phase encoding: AP, AP, PA, PA)

After Correction

---

✅ Protocol Checklist 4: Multiple b-values

Multi-shell scheme (HCP-like)

3 b-values: 1000 (20 directions), 2000 (40 directions), 3000 (60 directions)

Enables robust modeling across a range of diffusion strengths

Grid scheme (DSI-like)

23 b-values from 0 to 4000 s/mm², distributed across 258 directions

Optimized for q-space sampling and advanced reconstruction (e.g., DSI, GQI)

---

🧭 Session 3: Diffusion MRI Preprocessing (15 minutes)

Quality Issues in Diffusion MRI

Artifact	Cause	Consequence	Solution
motion artifacts	subject moves during the scan	signal dropout and between-dwi misalignment → hairball-like tractography	(1) discard the scan if motion is severe (2) apply motion correction to realign images
eddy current artifacts	gradient-induced eddy currents distort the readout	between-dwi misalignment → hairball-like tractography	(1) use current-canceling gradient designs (e.g., bipolar, twice-refocused) (2) apply affine image registration
susceptibility artifacts	magnetic field distortion near air-tissue interfaces	signal dropout and geometric distortion along the phase-encoding direction	(1) use sequence-based corrections (e.g., segmented EPI) (2) combine reversed-phase b0 images to correct distortion
flipped b-table	common in animal scans	urchin-like tractography	automatic b-table checking
rotated/flipped image volume	common in animal scans	cannot use atlas or autotract functions	flip or swap axis in pair
thick slices	old DWI sequence	poor fiber tracking result	regrid images

other corrections (less important): noise reduction, bias field correction, gibbs ringing correction

🛠️ Tools for Corrections

Popular Tools: ✅ FSL • ✅ MRtrix3 • ✅ QSIPrep • ✅ DIPY • ✅ DSI Studio

• FSL’s topup: corrects nonlinear distortion caused by susceptibility
• FSL’s eddy: corrects linear distortion from eddy currents & subject motion
• DSI Studio’s motion correction: corrects for eddy current and motion artifacts

---

🖐️ Hands-On Practice

Identify issues on [OpenNeuro ds002087] and correct it

Identify reversed phase encoding b0 for TOPUP

• dwi
• pa b0
• bval, bvec

---

🧩 Preprocessing Options

1. FSL’s topup + eddy → most comprehensive
2. FSL’s eddy only → for datasets lacking reverse-phase b0
3. DSI Studio motion correction → quick fix when others are unavailable

---

🖐️ Hands-On Practice

Compare correction results using Quality Control

1. Download .sz files from [Fiber Data Hub – ds002087][ds002087]
2. Use QC routine to compare .sz and .fz with/without correction
3. Check “diffusion contrast” and “Neighboring DWI correlation”

---

🧭 Session 4: Diffusion Modeling Methods (15 minutes)

---

---

🖐️ Hands-On Practice

Reconstruct DTI and GQI data and compare FA and QA

---

🧰 DTI:

• ⚠️ single tensor model cannot handle crossing fibers
• ⚠️ FA maps are unreliable at low SNR, yet often used in fiber tracking.
• ✅ still effective when paired with sufficient spatial resolution and smoothing

---

🧰 GQI / QSDR:

• ⚠️ Limited power for resolving complex crossing fibers
• ⚠️ sensitive to b1 inhomogniety
• ✅ main advantage: Provides a robust anisotropy index (QA) to guide fiber tracking

---

Trade-off between sensitivity and sepcificity

source: Yeh, Fang-Cheng, et al. “Tractography methods and findings in brain tumors and traumatic brain injury.” Neuroimage 245 (2021): 118651.

source: Kjer, Hans Martin, et al. “Bridging the 3D geometrical organisation of white matter pathways across anatomical length scales and species.” Elife 13 (2025): RP94917.

---

🧭 Session 5: GUI-based Batch Processing (10 minutes)

• NIFTI/bids OpenNeuro ds001378 (SCA2)

  1. run [NIFTI Quality Control]
  2. run [Step B2a: NIFTI to SRC (BIDS)]
  3. run [SRC Quality Control]
  4. SRC to FIB
  5. FIB to tractography

• DICOM: An MRI DICOM data set of the head of a normal male human aged 52

  1. Rename & Sort DICOM files
  2. Convert DICOM to NIFTI

Assignment 1: Explore Correction Effects on Tractography

1. Download .sz files from [Fiber Data Hub][Open Neuro][ds002087]
2. Visualize arcuate fasciculus with/without corrections and with/without head motion