A priori about setting excitability parameters η in combination with clinical hypothesis

[lala z]

Hello, 

        I want to ask if there is any basis for setting the prior parameters when combination with clinical hypothesis? Because it is mentioned in the paper that if the excitability parameter of a certain brain region ηc> - 2.05, this brain area is EZ, if ηc- Δη < η < ηc. Then the brain area is Pz, Δη= 0.4, that is to say, it belongs to PZ within the range of (-2.45, -2.05), while η < ηc- Δη This brain region belongs to Hz. In the literature, if a brain region is clinically assumed to be EZ, the prior is N (- 1.6,1), otherwise N (- 3,1). What is the basis for setting - 1.6 and - 3?

I also tried to combine clinical hypothesis, but it seems that the priori belief is too strong, and the results are shown in the following figure:

I use the form of reparameterized non central parameter in BVEP,  η=η_ mu+0.1*eigen_ vec* η_ star, η_ star~N (0,1), and I found that the predicted η>η_ mu, so if I combine the clinical hypothesis, even though the EZ of the clinical hypothesis is not the real EZ, if its prior mean value is set to η_ mu=- 1.6, then the predicted η  will be greater than - 1.6, so it will be meaningless if I combine  with clinical hypothesis.

[WOODMAN Michael]

hi

The basis for -1.6 and -3 are just based on observations of what works, but that's for datasets that we've tested.  You'll need to experiment a bit to see what works for your datasets.  There can be a lot of variability in sEEG data. 

cheers,

Marmaduke

---

From: 'lala z' via TVB Users <tvb-...@googlegroups.com>
Sent: Thursday, November 24, 2022 12:46:07 PM
To: TVB Users
Subject: [RESEAUX SOCIAUX] [TVB] A priori about setting excitability parameters η in combination with clinical hypothesis

 

--
You received this message because you are subscribed to the Google Groups "TVB Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tvb-users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/e4a45989-d30a-41c7-996f-f273b6d8d209n%40googlegroups.com.

[lala z]

Hello, 

        I have an idea. I don't know whether I think it is reasonable, because from my results, if I set a priori N (- 1.6,1) for the brain region with the clinical hypothesis of EZ, I think the priori belief is too strong. Maybe I can set the priori of the excitability parameters of the brain region with the clinical hypothesis of EZ within the range of PZ. If the observation value supports the priori, then those brain regions with the clinical hypothesis of EZ will be predicted to be EZ, If the observations do not support a priori, they will be predicted as PZ instead of EZ. I don't know whether this setting is reasonable?

Do you have any suggestions？

[WOODMAN Michael]

hi, 

Yes this is what I meant by experiment a little bit.  You may also want to weaken the priors by increasing their standard deviation e.g. N(-1.6, 2). 

cheers,

Marmaduke

---

From: 'lala z' via TVB Users <tvb-...@googlegroups.com>

Sent: Friday, November 25, 2022 9:04:57 AM
To: TVB Users
Subject: [RESEAUX SOCIAUX] Re: [RESEAUX SOCIAUX] [TVB] A priori about setting excitability parameters η in combination with clinical hypothesis

 

Hello, 

        I have an idea. I don't know whether I think it is reasonable, because from my results, if I set a priori N (- 1.6,1) for the brain region with the clinical hypothesis of EZ, I think the priori belief is too strong. Maybe I can set the priori of the excitability parameters of the brain region with the clinical hypothesis of EZ within the range of PZ. If the observation value supports the priori, then those brain regions with the clinical hypothesis of EZ will be predicted to be EZ, If the observations do not support a priori, they will be predicted as PZ instead of EZ. I don't know whether this setting is reasonable?

在2022年11月24日星期四 UTC+8 21:27:16<marmaduke.woodman> 写道：

To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/998fe13f-5b43-4234-9f9c-4462ba316545n%40googlegroups.com.

[lala z]

Hello,

       the prior I set is to use the reparameterization method in the BVEP code, that is, to introduce the gain matrix, eta=eta_ mu+0.1*eigen_ vec*eta_ star , eigen_ vec=Gain.T×Gain , eta_ star~N (0,1), in fact, I don't understand why this setting is required, but it is a fact that the result of this setting is better than that of directly setting eta=eta_ mu+eta_ star,eta_ star~N (0,1). How can I weaken the  priors  in this case? I tried to drop 0.1, but it seems that the result is not very good. Should I adjust eta_ star~(0,2)?

[lala z]

Hello, I have a few questions to ask you:
1. I read that the excitability parameter η is set as a non-informative prior in the paper,  N (- 3,1).   Is "- 3"  the most appropriate value obtained through some experiments? For my own data, if I set the prior to N (- 3,1), most brain regions will be predicted to be PZ or HZ. Do I need to adjust the mean of the prior?

2. In combination with the clinical hypothesis, I don't know how to judge whether the parameters I set are the most appropriate, that is, the priori belief can't be too strong. I think of a method to verify, that is, arbitrarily set several brain nodes, treat them as EZ, and set a priori belonging to EZ to verify whether they will be predicted as EZ. If they are predicted as EZ, it means that the parameter priori belief I set is too strong, Otherwise, it is considered that the parameters I set are appropriate. Do you think this is reasonable?
Thanks!

lala

[lala z]

Hello, 

         I have encountered a very strange problem, that is, from all my results, all the finally predicted epileptic parameters η should be higher than the prior mean value I set, for example, when I set η=η_ mu+0.1*eigen_  vec*  η_  star,  η_  Star~N (0,1), if η_ mu=- 3, then finally predicted η  of the posterior distribution  should be greater than - 3, which will lead to too strong a priori belief. Therefore, if I combine the clinical hypothesis, it will be meaningless. Is this a normal phenomenon? My results are shown in the figure below. It can be said that all my results are like this:

It can be seen from this figure that the posterior distribution of all parameters moves upward as a whole and is higher than the mean value of the prior distribution（ η_ mu=-3）,Therefore, if I set the clinical hypothesis EZ a prior distribution with a large mean value（η_ mu=-1.6 or -2.25 >-3）, then the predicted posterior of these nodes will also be higher than the mean value of the prior distribution, then it seems to have little significance.

在2022年11月25日星期五 UTC+8 17:28:02<marmaduke.woodman> 写道：

[WOODMAN Michael]

hi again,

I agree this looks like the model is overconstrained.  In the model code

n = n_mu + 0.1 * eigen_vec * n_star

you may wish to treat 0.1 as a hyper parameter, e.g.

data { 

   ....

   real n_scale;

}

and then use

n = n_mu + n_scale * eigen_vec * n_star

and try different values of scale.  

cheers,

Marmaduke

---

From: 'lala z' via TVB Users <tvb-...@googlegroups.com>

Sent: Wednesday, November 30, 2022 4:21:25 AM
To: TVB Users
Subject: [RESEAUX SOCIAUX] Re: [RESEAUX SOCIAUX] Re: [RESEAUX SOCIAUX] [TVB] A priori about setting excitability parameters η in combination with clinical hypothesis

 

To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/7a9b1268-bdc3-49a7-9817-ebbb3fe78823n%40googlegroups.com.