🧭 Workshop Week 3 Outline

1. Fiber Tracking Algorithm
   • Principles and implementation of deterministic and probabilistic tracking
   • Tracking parameters and termination criteria
2. Track Filtering Using Regions
   • Role of ROI, ROA, END, and other region types in shaping tractography results

3. AutoTrack

4. Connectomics
   • Region-to-region connectome
   • Tract-to-region connectome as a workaround for crossing/kissing ambiguity

---

🧠 Session 1: Fiber Tracking Algorithm

🔹 Input Requirements

• Local fiber orientations (also known as fixels)
• Anisotropy

Sample Dataset: [OpenNeuro][ds004299][sub-103_ses-1_dwi.gqi.fz]

🔹 Tracking Steps

1. Select a starting point in the seed region and an initial direction
2. Repeat the following loop:
    2.1 Check termination conditions (e.g., low anisotropy, sharp turning angle)
    2.2 If conditions are met, proceed to Step 3
    2.3 Determine a new propagation direction based on local fiber orientation and prior path
    2.4 Move one step forward
3. If only one direction has been tracked, return to Step 1, reverse the initial direction, and repeat tracking to generate the full streamline.
    Otherwise, tracking ends.

---

⚙️ Tracking Parameters

• Anisotropy threshold – stops tracking in low-contrast regions
• Angular threshold – stop at sharp turns
• Step size – distance advanced per iteration
• Minimum/Maximum length – defines acceptable tract lengths
• Maximum seeds/tracts – limits total output
• Track-to-voxel ratio

Key Challenge in Tractography: Crossing vs Kissing Configuration

---

🎯 Deterministic vs. Probabilistic Fiber Tracking

🔵 Deterministic Tracking

• At Step 2.3, always follows the direction with the smallest turning angle
• Assumes all resolved orientations represent crossing fibers
• Direction is chosen from local maxima of the GQI ODF

📌 Example:

• GQI + deterministic tracking

⚠️ Limitations (False Negatives):

• May misinterpret large turns as crossings → early termination
• May misinterpret sharp crossings as turns → incorrect continuation

---

🔴 Probabilistic Tracking

• At Step 2.3, selects propagation direction based on a probability distribution
• Resolves multiple orientations due to crossing or turning
• Fiber configuration is inferred from accumulated probability over many iterations

📌 Examples:

• bedpostx + probtrackx
• CSD + iFOD2

⚠️ Limitations (False Positives):

• May produce spurious pathways due to false configurations

🧬 Histology Reference:
Allen Brain Atlas – Parvalbumin Stain

---

🧩 Session 2: Track Filtering Using Regions

🗂️ Region Types

1. Seed – Region where tracking starts
2. ROI (Region of Interest) – Tracts must pass through; otherwise, they are discarded
3. ROA (Region of Avoidance) – Tracts that enter this region are discarded
4. Limiting – Tracts that exit this region are discarded
5. END – Tracts that do not terminate within this region are discarded
6. NotEND – Tracts that terminate within this region are discarded
7. Terminating – Tracts are forced to stop upon entering this region

---

💡 Rule of Thumb

• ✅ Prefer whole brain seeding unless you are confident certain regions should be excluded from the tract
• ⚠️ Avoid using END regions too early — test with ROI first to make sure the tract reaches the target without premature termination

---

🔍 Common Use Cases

📌 To find connections of a single region:

• One ROI + whole brain seeding
• One ROI + dilated tract coordinates used as Seed + Limiting

📌 To find connections between two regions:

• Two ROIs + whole brain seeding
• Two ROIs + dilated tract coordinates as Seed + Limiting
• One ROI + one END + dilated tract coordinates as Seed + Limiting
• Two ENDs + dilated tract coordinates as Seed + Limiting

---

Session 3: AutoTrack

Sample Data: [Other major studies][penthera]

• [sub-PT001_ses-1_dwi.gqi.fz]
• [sub-PT001_ses-2_dwi.gqi.fz]
• [sub-PT001_ses-3_dwi.gqi.fz]

---

🧠 Session 4: Connectome

🔗 Region-to-Region (R2R) Connectome

📄 Yeh, Fang-Cheng, et al. NeuroImage, 2018

Workflow:

• Run whole-brain tractography (e.g., >1 million streamlines)
• Apply a brain parcellation scheme
• Count the number of tracts connecting each pair of regions

Limitations:

• Tract count lacks biological specificity
• Ground truth is unattainable due to crossing/kissing ambiguities
• Limited sensitivity to brain disorders

---

🧬 Tract-to-Region (T2R) Connectome

📄 Yeh, NeuroImage, 2020

Workflow:

• Use autoTrack to extract individual tract bundles (test-retest validated)
• Apply a brain parcellation
• Compute the volume fraction of each region that is traversed by the tract → defines T2R connectivity

Advantages over R2R:

• Produces physically interpretable metrics (e.g., % of region volume)
• Reduces uncertainty from crossing/kissing ambiguities by evaluating bundles individually

Example

---

Assignment: Plot T2R connectome on HCP-MMP parcellation

• At [Fiber Data Hub], select
  • a baby in the [HCP Lifespan][BCP] study (scanned at 3 months, 6 months, and 2 years)
  • a child from [OpenNeuro][ds003604] scanned at 5, 7, and 9 years)
  • a teenager from the [ABCD] study (scanned at 10, 12, and 14 years)
• compute their t2r connectome of the arcuate fasciculus and HCP-MMP parcellation
• visualize the t2r connectome using region rendering