Controlling for site / lesions in TBI studies

[Finian Keleher]

Hi Frank,

I’m running correlational and differential tractography on two longitudinal pediatric TBI datasets, and I have a couple of questions I’d really appreciate your input on:

1. Both datasets were collected across multiple sites with different scanner parameters. For correlational tractography, is there a way to account for site effects? Similarly, for differential tractography, I’ve already generated loss/gain maps and TDI for all participants, but I’m hesitant to proceed with group-level comparisons given these site differences. Do you have a recommended way to control for site in this context?
2. One dataset includes moderate-to-severe TBI patients, many of whom have lesions. Is there a way to incorporate lesion masks into correlational or differential tractography? Or can either of these approaches handle lesions? What would you recommend?

Thanks very much for your help.

[Frank Yeh]

Hi Finian,

Regarding your questions about controlling for site and handling lesions:

To account for site effects in both correlational and differential tractography, you can add binary variables (0 and 1) for each site, similar to how you would code for gender. For example, if you have three sites, you should include three separate variables to represent each group.

As for incorporating lesion masks, my experience suggests that it is generally better to treat the data as you would for a healthy subject. Specifically incorporating lesion masks can feed the expected results back into the analysis, which risks circular reasoning.

I hope this helps.

Best regards,

Frank Yeh

[Finian Keleher]

Thanks for your response, Frank!

For the patients with lesions, do you think leaving lesions unmasked could introduce issues at later stages, particularly during registration to template space or when averaging across subjects for group comparisons? I’m especially concerned about whether lesions might introduce bias or distort group-level results.

Would you recommend proceeding with future steps as I would for healthy subjects, or are there any approaches you would suggest when working with patients with lesions in this context?

[Frank Yeh]

For the patients with lesions, do you think leaving lesions unmasked could introduce issues at later stages, particularly during registration to template space or when averaging across subjects for group comparisons? I’m especially concerned about whether lesions might introduce bias or distort group-level results.

My experiences were the registration is good with that.
 

Would you recommend proceeding with future steps as I would for healthy subjects, or are there any approaches you would suggest when working with patients with lesions in this context?

Yes, the point here is that, especially for testing a hypothesis, identical analysis procedures should be applied to subject's and patients in order to have a fair comparison, otherwise, there is a risk of circular logic, and we may getting results/bias due to lesion we added.

Best,

Frank

 

 

On Friday, May 1, 2026 at 6:39:18 AM UTC-6 Frank Yeh wrote:

Hi Finian,

Regarding your questions about controlling for site and handling lesions:

To account for site effects in both correlational and differential tractography, you can add binary variables (0 and 1) for each site, similar to how you would code for gender. For example, if you have three sites, you should include three separate variables to represent each group.

As for incorporating lesion masks, my experience suggests that it is generally better to treat the data as you would for a healthy subject. Specifically incorporating lesion masks can feed the expected results back into the analysis, which risks circular reasoning.

I hope this helps.

Best regards,

Frank Yeh

--
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/6e20d346-3c0b-4bbc-a153-9aa7e592d478n%40googlegroups.com.