Registration/Normalization and Statistical Analysis
Maryam Shapourjani
Hello Dear Frank,
My goal is to model the growth of shape metrics (e.g., span, volume, etc.) related to dHCP neonate brain tracts concerning increasing age (37 to 44 weeks).
After using the GQI reconstruction method (to avoid changing the shape metrics) and then performing automated tractography, I extracted the statistics.
Now I have some questions:
1. GQI ReconstructionIs it correct that in GQI reconstruction, the reconstruction is performed in the subject's local space?
A. In GQI reconstruction, is there no normalization applied to the subject space?
B. In GQI reconstruction, is no registration (neither linear nor non-linear) used on the subject space?
2. Automated Tractography
Is it correct that in the automated tractography stage, the tractography is performed in the subject space?
A. In the automated tractography stage, is there no normalization applied to the subject space?
B. In the automated tractography stage, is the subject space not subjected to any registration (neither linear nor non-linear)?
3. Registration/Normalization and Shape Metrics
If, in one of the stages (GQI reconstruction or automated tractography), the subject space undergoes registration or normalization, will this registration or normalization not cause a change in the shape metrics?
A. If the answer is no (i.e., it does change the metrics), what is the reason?
4. Statistical Analysis and Modeling (Assuming No Registration/Normalization)
If, in the GQI reconstruction and automated tractography stages, the subject space remains without applied registration or normalization, a question arises about the statistical analysis and modeling of the growth of shape metrics (e.g., span, volume, etc.):
A. Considering that the brain of each subject is unique in terms of size, shape, and anatomical folding, will the statistical analysis and modeling of brain growth not yield weak statistical results and invalid conclusions?
B. If the answer to the previous question is that invalid results are obtained, what solution do you propose to resolve this issue? A solution that neither changes the shape metrics nor results in invalid conclusions for the analysis.
Maryam Shapourjani
Frank Yeh
> I want to examine the growth of a tract in terms of its **length** as the neonate ages. For example, this tract should extend from the **temporal** lobe to the **frontal** lobe as it grows.
> Let's assume that when examining it on a template, the tract's length is **10 millimeters (mm)** at the time it reaches from the temporal to the frontal lobe.
>
> Now, consider the length of this same pathway in the **subject's native space** that I am examining:
> neonates have different head sizes. Suppose there is a neonate with a **small head** and a neonate with a **large head**, and both have been tractography in their native space. I intend to examine the growth of the pathway that reaches from the temporal to the frontal lobe in both neonates.
> When examining in the native space of the two neonates, it is possible that:
> * In the **neonate with the smaller head**, the tract has reached from the temporal to the frontal lobe, but its length is **not 10 mm** and is **shorter (e.g., 6 mm) **.
> * In the **neonate with the larger head**, the length of the same tract is **10 mm**, but its growth is **not yet complete**, and it has **not reached from the temporal to the frontal lobe**.
>
> As I said in the example above, if the tract length is examined in the subject's native space, it is affected by **brain size**.
Logically, I may argue the brain size is affected by tract length, not
the tract length affected by brain size. In the neurodevelopmental
aspect, brain size is the final result.
> Considering that I am modeling tract growth based on **tract shape metrics**, and length is a shape metric, how can it be argued that examining growth in the subject's native space does **not** cause errors in growth assessment and **does not** lead to incorrect analysis results?
Having covariates attributing correlation does not necessarily mean it
is "incorrect" or "invalid"
This is why I asked to define "invalid" first. It can mean the model
assumption of independence is violated.
I am not sure if having covariates affecting correlation results is
considered to be "invalid."
>
> **Conclusion: **
> Since I am studying neonates, I want the **length to be dependent on growth** and **not dependent on brain size**.
> In other words, changes in brain size should **not** be considered as growth; rather, the greater the length, the more it indicates a closer proximity to the **end-point**.
>
This is a statistical modeling topic, and I would highly recommend
checking them out with an expert biostatistician.
Best regards,
Frank
> Based on the descriptions above, I would appreciate your guidance on question 4.
>
> Sincerely,
> Maryam Shapourjani
> On Sat, 22 Nov 2025 at 17:36, Frank Yeh <fran...@gmail.com> wrote:
>>
>> > 1. GQI Reconstruction
>> >
>> > Is it correct that in GQI reconstruction, the reconstruction is performed in the subject's local space?
>> > A. In GQI reconstruction, is there no normalization applied to the subject space?
>> > B. In GQI reconstruction, is no registration (neither linear nor non-linear) used on the subject space?
>> > ________________________________
>> >
>> > 2. Automated Tractography
>> >
>> > Is it correct that in the automated tractography stage, the tractography is performed in the subject space?
>> > A. In the automated tractography stage, is there no normalization applied to the subject space?
>> > B. In the automated tractography stage, is the subject space not subjected to any registration (neither linear nor non-linear)?
>> > ________________________________
>> >
>> > 3. Registration/Normalization and Shape Metrics
>> >
>> > If, in one of the stages (GQI reconstruction or automated tractography), the subject space undergoes registration or normalization, will this registration or normalization not cause a change in the shape metrics?
>> >
>> > ________________________________
>> >
>> > 4. Statistical Analysis and Modeling (Assuming No Registration/Normalization)
>> >
>> > If, in the GQI reconstruction and automated tractography stages, the subject space remains without applied registration or normalization, a question arises about the statistical analysis and modeling of the growth of shape metrics (e.g., span, volume, etc.):
>> >
>> > A. Considering that the brain of each subject is unique in terms of size, shape, and anatomical folding, will the statistical analysis and modeling of brain growth not yield weak statistical results and invalid conclusions?
>>
>> >
>> > B. If the answer to the previous question is that invalid results are obtained, what solution do you propose to resolve this issue? A solution that neither changes the shape metrics nor results in invalid conclusions for the analysis.
>> >
>> >
>> > Sincerely,
>> > Maryam Shapourjani
>> >
>> > 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/43e96e35-89cf-4933-94b6-6d8633290a45n%40googlegroups.com.