Optimal QA Threshold Strategy for ADNI/PPMI Group Study (Auto vs Fixed?)

[Jaeyoung Kang]

Hi Frank,

I am processing a large group study (ADNI/PPMI data) using DSI Studio (command line via Docker). My current pipeline is:

1. Reconstruction: QSDR (Human Connectome Template)
2. Tracking: Whole-brain seeding, 1M seeds, threshold_index=qa.
3. Connectivity: Brainnetome atlas to generate connectivity matrices.

The Issue: I am currently leaving the threshold value unspecified (so it uses the Auto/Default threshold). However, when analyzing the resulting binary connectivity matrices, I am seeing cutoff regions / disconnected nodes (regions with 0 connections) in many subjects. Ideally, I want to minimize these false negatives for a valid network analysis.

My Questions:

1. Auto vs. Fixed: For a large heterogeneity dataset like ADNI/PPMI, is it safe to rely on the auto-threshold (Otsu-based)? Or is it standard practice to set a conservative fixed lower bound (e.g., fa_threshold=0.15 or 0.1) to ensure consistency across the group?
2. Individual Threshold: If "Auto" is inconsistent, is there a recommended way to determine the optimal "individual" threshold for each subject if I script it myself, or is a fixed group threshold preferred?

Thank you so much for your help!

Jae

[Frank Yeh]

Sorry for delayed reply and this is a challenging task and I would say there's no correct or the best solution. The best setting depends on your data set and I will recommend you test different settings to see which one hass the best reproducibility.

The most updated version removes any thresholding so the matrix are not threshold. The purpose is to let users apply their own thresholding.

Best regards

Frank

--
You received this message because you are subscribed to the Google Groups "DSI Studio" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dsi-studio+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/dsi-studio/7011fb07-327a-4636-8941-331775a489ffn%40googlegroups.com.