Re: 回复： [SUSPECTED SPAM] [TVB] Process Empirical SEEG data

[WOODMAN Michael]

hi,

Please also look at the estimated offsets and amplitude parameters, since these might explain what you're seeing.  

cheers,

Marmaduke

---

From: '啦啦啦' via TVB Users <tvb-...@googlegroups.com>
Sent: Tuesday, July 19, 2022 9:12:40 AM
To: tvb-users
Subject: 回复： [SUSPECTED SPAM] [TVB] Process Empirical SEEG data

 

Hi,

sorry to bother you again. I'm using our own empirical SEEG data to infer the value of eta parameters with stan, but I do encounter some problems and don't know how to solve them.

My SEEG data and log power are shown in the figure:

I ran the BVEP/stan/BVEP1_ seeg_ noncentered_ reparam.stan, and some parts are modified as shown in the figure:

I added the "sqrt(dt)*x_eta[i, t]*sig,  sqrt(dt)*z_eta[i, t]*sig" ,which is not in the original file, and set sigma=0.16,eps=0.06.

Only in this way can the fitted SEEG achieve a more ideal effect,show in the figure:

However, the posterior estimation of eta seems not very good, as shown in the figure:

It can be seen from the figure that SEEG data seems to have no effect on inferring the posterior distribution of eta, I try to change the prior distribution of eta, for example,(eta= -2.5+0.1*eigen_vec*eta_star) or (eta= -3.5+0.1*eigen_vec*eta_star).

Is it related to the gain matrix? Maybe I should adjust other parameters?

I saw that it was mentioned to use the openmeeg package to calculate the gain matrix. Will it be helpful?

I would appreciate it if you could give me some advice!

lala

------------------ 原始邮件 ------------------

发件人: "tvb-users" <marmaduk...@univ-amu.fr>;

发送时间: 2022年6月27日(星期一) 下午4:47

收件人: "tvb-...@googlegroups.com"<tvb-...@googlegroups.com>;

主题: Re: [SUSPECTED SPAM] [TVB] Process Empirical SEEG data

hi,

There could be a seizure there but hard to say, we'd need to see the raw time series.  Unfortunately, please do not share figures of patient data on this public mailing list.  Please contact us privately for a quick call to see if there's a quick fix.

cheers,

Marmaduke

---

From: 'lala z' via TVB Users <tvb-...@googlegroups.com>
Sent: Thursday, June 23, 2022 4:31:50 AM
To: TVB Users
Subject: [SUSPECTED SPAM] [TVB] Process Empirical SEEG data

 

Hello,

         I am now trying to use real SEEG data for fitting. I intercepted the SEEG data of seizures marked by clinicians，and then processed the SEEG data according to the methods provided by you, but the data used for fitting is different from yours, as shown in the following figure：

Is this correct? Maybe there are other requirements for using real SEEG data？

Thanks,

lala

--
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/349743ca-faf7-4bc0-9faf-e9ad51a53592n%40googlegroups.com.

--
You received this message because you are subscribed to a topic in the Google Groups "TVB Users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/tvb-users/xO9rmNJDZtk/unsubscribe.
To unsubscribe from this group and all its topics, send an email to tvb-users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/08ed4ddd2f2d47e48018c1eeb09ea6da%40univ-amu.fr.

--
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/tencent_1809A1F5E2A665069DFECCCD85CAE577600A%40qq.com.

[啦啦啦]

So is it better to calculate the gain matrix with openmeeg?

------------------ 原始邮件 ------------------

发件人: "tvb-users" <marmaduk...@univ-amu.fr>;

发送时间: 2022年8月3日(星期三) 凌晨1:04

收件人: "tvb-...@googlegroups.com"<tvb-...@googlegroups.com>;

主题: Re: 回复： [SUSPECTED SPAM] [TVB] Process Empirical SEEG data

To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/f51c11e20bdc4e3d9428cf99ece11974%40univ-amu.fr.

[啦啦啦]

Hi,

And now I have another problem, that is, I use the same parameters to run the code twice, and the results are different. The results of the two runs are shown in the figure:

And the results of these two times are quite different. What is the reason? Do you have any suggestions?

lala

------------------ 原始邮件 ------------------

发件人: "tvb-users" <tvb-...@googlegroups.com>;

发送时间: 2022年8月3日(星期三) 下午3:03

收件人: "tvb-users"<tvb-...@googlegroups.com>;

主题: 回复： 回复： [SUSPECTED SPAM] [TVB] Process Empirical SEEG data

To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/tencent_614FE39CB15C4487A1D046EF6F1A5950E406%40qq.com.

[WOODMAN Michael]

Hi,

The general trends are consistent across runs is similar.  Given how these algorithms work it's possible for runs to be a bit different depending on the initialization. 

cheers,

Marmaduke

---

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

Sent: Wednesday, August 3, 2022 12:59:49 PM
To: tvb-users
Subject: 回复： 回复： [SUSPECTED SPAM] [TVB] Process Empirical SEEG data

 

To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/tencent_E13FA473A572341D414292BDC4696941DE08%40qq.com.

[啦啦啦]

Hi,

thank you for your reply. Is there any way to improve it?  Although the general trends are consistent across runs is similar, it is impossible to accurately estimate EZ in this way, because I don't know which operation result is reliable.

------------------ 原始邮件 ------------------

发件人: "tvb-users" <marmaduk...@univ-amu.fr>;

发送时间: 2022年8月3日(星期三) 晚上10:33

收件人: "tvb-...@googlegroups.com"<tvb-...@googlegroups.com>;

主题: Re: 回复： 回复： [SUSPECTED SPAM] [TVB] Process Empirical SEEG data

To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/cfa1f68253f14f668758d981ef32308b%40univ-amu.fr.

[WOODMAN Michael]

What algorithm (advi, hmc, etc) are you using?

---

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

Sent: Thursday, August 4, 2022 08:52
To: tvb-users
Subject: 回复： 回复： 回复： [SUSPECTED SPAM] [TVB] Process Empirical SEEG data

 

Hi,

[啦啦啦]

The algorithm I use is advi

------------------ 原始邮件 ------------------

发件人: "tvb-users" <marmaduk...@univ-amu.fr>;

发送时间: 2022年8月8日(星期一) 下午5:54

收件人: "tvb-...@googlegroups.com"<tvb-...@googlegroups.com>;

主题: Re: 回复： 回复： 回复： [SUSPECTED SPAM] [TVB] Process Empirical SEEG data

--
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/1836d77bb5a144ba86ca54e82b6ce72a%40univ-amu.fr.

[WOODMAN Michael]

With ADVI, you will have more stable results when using many MC samples for the ELBO estimate.  Are you using Stan or PyMC3?

---

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

Sent: Monday, August 8, 2022 12:23:51 PM
To: tvb-users
Subject: [RESEAUX SOCIAUX] 回复： 回复： 回复： 回复： [SUSPECTED SPAM] [TVB] Process Empirical SEEG data

 

To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/tencent_069D743408D9F68EA5650769FC80CD7DCA0A%40qq.com.

[啦啦啦]

I use Stan. What do many MC samples mean? Do I need more electrodes or more iterations?

------------------ 原始邮件 ------------------

发件人: "tvb-users" <marmaduk...@univ-amu.fr>;

发送时间: 2022年8月8日(星期一) 晚上6:34

收件人: "tvb-...@googlegroups.com"<tvb-...@googlegroups.com>;

主题: Re: [RESEAUX SOCIAUX] 回复： 回复： 回复： 回复： [SUSPECTED SPAM] [TVB] Process Empirical SEEG data

To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/3da775ab6bc44c73a225a0d088185840%40univ-amu.fr.

[WOODMAN Michael]

hi,

It's explained in the docs here 

https://mc-stan.org/docs/2_24/cmdstan-guide/variational-inference-algorithm-advi.html

you can set the grad_samples to a number greater than 1 for a more stable estimate. I would recommend e.g. 10 to start with. 

cheers,

Marmaduke

---

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

Sent: Monday, August 8, 2022 12:40:05 PM
To: tvb-users
Subject: [RESEAUX SOCIAUX] 回复： [RESEAUX SOCIAUX] 回复： 回复： 回复： 回复： [SUSPECTED SPAM] [TVB] Process Empirical SEEG data

 

To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/tencent_9D4550C5CFD85CA7DC5EC7A950476F7A1A06%40qq.com.

[啦啦啦]

OK, thank you very much for your help

------------------ 原始邮件 ------------------

发件人: "tvb-users" <marmaduk...@univ-amu.fr>;

发送时间: 2022年8月8日(星期一) 晚上7:40

收件人: "tvb-...@googlegroups.com"<tvb-...@googlegroups.com>;

主题: Re: [RESEAUX SOCIAUX] 回复： [RESEAUX SOCIAUX] 回复： 回复： 回复： 回复： [SUSPECTED SPAM] [TVB] Process Empirical SEEG data

To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/93d7a99829d743f6a9cdea9f280db351%40univ-amu.fr.

[啦啦啦]

Hello, 

I increased the value of grad_sample to 15, but the result still seems to be not very stable, and the larger the value of grad_sample, the longer the running time. Maybe I need to adjust other parameters or continue to increase the value of grad_sample?

In addition, the accuracy of EZ zone prediction with empirical seeg data is also relatively low, and the results are shown in the figure:

The above figure shows two different results of running with the same parameters. Since there is no PZ hypothesis in our clinical data, I only classified Ez and Hz.

Therefore, there are two problems: 1. The result is still unstable; 2. The accuracy of EZ prediction with empirical SEEG data is relatively low. I guess that the reason for the low accuracy of EZ prediction may be the SEEG data and the gain matrix. I can replace the SEEG data, but is there a better way to calculate the gain matrix?

Do you have any other suggestions?

Thanks!

lala

[WOODMAN Michael]

hi

Do you have an image of the gain matrix or some statistics on its values?  If you have an image of the positions of the sensors relative to the positions of the regions? It may be that it is not well conditioned causing issues for estimations.

cheers,

Marmaduke

[lala z]

Hello, I seem to find that the accuracy of EZ prediction is very low, mainly because the brain area node that contributes the most to the seeg electrode in the gain matrix is inconsistent with the anatomical position of the seeg electrode in the brain segmentation map. The gain matrix is calculated from the coordinates of the brain area center and the seeg electrode, and the anatomical position of seeg in the brain segmentation map is determined according to the voxel position and electrode position in the brain segmentation map, Therefore, there may be errors. Is my guess correct? What can be done to improve it?