{ "id": 10254, "title": "Electrode name to location mappings", "url": "/competitions/seizure-prediction/discussion/10254", "author_name": "", "post_date": "2014-09-07T21:11:59.710Z", "votes": null, "comment_count": 19, "views": 4698, "content": "

Hi all,

\n

Has any found or could AES/Kaggle provide mappings from the electrode names to the electrode placement on the brain?

\n

I cannot find a method to deconvolve the apparently three different naming standards (Dog_1-5, Patient_1 and Patient_2).

\n

Patient_1: 16 electrodes named LD_1, ..., LD_8, RD_1, ..., RD_8

\n

Patient_2: 24 electrodes named LGT_01, ..., LGT_24

\n

Dog_1: 16 electrodes named NVC1202_32_002_Ecog_c001, ..., NVC1202_32_002_Ecog_c016

Dog_2: 16 electrodes named NVC0905_22_002_Ecog_c001, ..., NVC0905_22_002_Ecog_c016

Dog_3: 16 electrodes named NVC0906_22_007_Ecog_c001, ..., NVC0906_22_007_Ecog_c016

Dog_4: 16 electrodes named NVC1202_26_003_Ecog_c001, ..., NVC1202_26_003_Ecog_c016

Dog_5: 16 electrodes named NVC0905_22_004_Ecog_c001, ..., NVC0905_22_004_Ecog_c016

", "messages": [ { "id": "53306", "postDate": "09/07/2014 21:11:59", "content": "

Hi all,

\n

Has any found or could AES/Kaggle provide mappings from the electrode names to the electrode placement on the brain?

\n

I cannot find a method to deconvolve the apparently three different naming standards (Dog_1-5, Patient_1 and Patient_2).

\n

Patient_1: 16 electrodes named LD_1, ..., LD_8, RD_1, ..., RD_8

\n

Patient_2: 24 electrodes named LGT_01, ..., LGT_24

\n

Dog_1: 16 electrodes named NVC1202_32_002_Ecog_c001, ..., NVC1202_32_002_Ecog_c016

Dog_2: 16 electrodes named NVC0905_22_002_Ecog_c001, ..., NVC0905_22_002_Ecog_c016

Dog_3: 16 electrodes named NVC0906_22_007_Ecog_c001, ..., NVC0906_22_007_Ecog_c016

Dog_4: 16 electrodes named NVC1202_26_003_Ecog_c001, ..., NVC1202_26_003_Ecog_c016

Dog_5: 16 electrodes named NVC0905_22_004_Ecog_c001, ..., NVC0905_22_004_Ecog_c016

", "rawMarkdown": "", "votes": null }, { "id": "53324", "postDate": "09/08/2014 05:57:49", "content": "

I would also like to know the spatial map of the electrodes and the nominal voltage values for each subject (so we can remove spikes)

", "rawMarkdown": "", "votes": null }, { "id": "53394", "postDate": "09/09/2014 14:11:02", "content": "

This is the same situation as the previous seizure detection challenge (https://www.kaggle.com/c/seizure-detection/forums/t/9354/channel-name-to-location-of-electrode). It really isn't possible to provide an exact electrode location for this contest. For the dogs, the electrodes are positioned horizontally with two 4-contact strips on each side of the brain. Numbering goes from anterior to posterior, starting on the left superior strip. For the humans, there really is no standard system. Electordes are numbered on subdural grids, but grids can be inserted in any orientation. The best one can glean from the naming system is general anatomic coverage.

", "rawMarkdown": "", "votes": null }, { "id": "53397", "postDate": "09/09/2014 15:07:48", "content": "

Just to be clear: this means that channel 1 on one patient could be channel 4 on another patient (or not even exist).

", "rawMarkdown": "", "votes": null }, { "id": "53451", "postDate": "09/10/2014 16:33:26", "content": "

Yes. That's correct in patients. For the dogs in the contest the arrangement of electrodes is reasonably consistent. For human subjects the placement of electrodes is dictated by the clinical problem and the suspected location of seizure onset, which of course varies from patient to patient.

", "rawMarkdown": "", "votes": null }, { "id": "55195", "postDate": "09/25/2014 00:15:29", "content": "

A related question: On a single subject, are the electrode locations and identifications consistent over all examples?

", "rawMarkdown": "", "votes": null }, { "id": "55287", "postDate": "09/26/2014 15:10:31", "content": "

Yes. For each subject, each named iEEG channel records activity from a constant location in the brain throughout the entire recording. No electrodes were moved or replaced during the recording, nor were channel names changed.

", "rawMarkdown": "", "votes": null }, { "id": "55412", "postDate": "09/29/2014 22:17:14", "content": "

Just to clarify.

\n

In the recordings for the dogs, one could sort of assume that channel 1 is frontal, channel 4 is occipital and channels 2 and 3 could be like parieto-occipital?

\n

Also for humans, it could be very useful to know the general localization. Some phenomena are observable in the front cortex with eye movement (like Bell's) or some specific rithms by region, like alfa in the ccipital rebion and beta in the frontal one. If the data from patients is from grids and deep electrodes it could also be helpful to have that kind of data. I know mapping grids is difficult, but knowing if it is temporal or not could also help. Generally when neurosurgeons and epileptologists place grids, they make a small map of it. I wonder if this map is available.

\n\n

Thanks

\n\n

AG

", "rawMarkdown": "", "votes": null }, { "id": "55413", "postDate": "09/29/2014 22:34:30", "content": "

Also,

\n\n

Going back to the dogs, if there are 2 stripes with 4 electrodes from frontal to occipital numbered 1 to 4 per hemisphere of the brain. Which 2 electrodes are frontal left and which ones are frontal right? The reason for asking this is that probably the electrodes that are close together may see the same electrical activity with different intensity (e.g: the focci is between 2 electrodes).

\n\n

Thanks for the clarification

\n\n

AG

", "rawMarkdown": "", "votes": null }, { "id": "55414", "postDate": "09/29/2014 23:05:27", "content": "

AndresGarcia -

\n

Correct me if I am wrong, but I think this info is useless in this particular competition, because the model hast to be electrodes- and dog/patient- invariant, according to the rules. 

", "rawMarkdown": "", "votes": null }, { "id": "55415", "postDate": "09/29/2014 23:23:13", "content": "

Well, I guess it depends on how are you creating the algorithms. In a sense, one could shoot for a high sensitivity algorithm to include certain activity that looks suspicious of being preictal and then use another algorithm with high specificity to improve your probabilities of correctly finding a spike that will indicate abnormal electrical activity.

\n

I agree, the model has to be dog/patient invariant, but electrodes matter. When reading EEGs (in a usual setting) many times spikes of bipolar behavior are detected. This is that in one electrode the spike looks negative, while it looks positive in the electrode next to it. Hence, the electrical abnormal activity is happening between both electrodes. Sometimes preictal states are missed because misreading such spikes.

\n

Sometimes also, there is a spike in one side of the brain and then electrical activity in the contralateral part, meaning that there is spread (not contiguously but to the contralaeral hemisphere). In order for a good algorith to exists, one could shoot for groing reading by reading by electrode, but also to correlate such readings with other electrodes to increase specificity. I am not sure it is irrelevant...

\n\n

AG

", "rawMarkdown": "", "votes": null }, { "id": "55440", "postDate": "09/30/2014 20:24:19", "content": "

This may or may not  be helpful, but I'm posting with this email a diagram of the implanted electrodes in the dogs. This applies to all the dogs. As has been noted elsewhere Dog 5 had a channel that went bad, so we omitted it. Bear in mind that these locations are not exact - they represent targeted locations, and there could be quite a bit of variation in exactly where the surgeon placed the electrodes. We do not have post-implant imaging to verify electrode locations.

\n

As for the patient electrode maps and configurations, this is a bit more difficult, and I don't believe I will be able to provide much. Patients do typically have post-implant CT imaging, but I don't believe I can post any of that data here. I may however be able to give a general description of how the electrodes are arranged - I will check on that, both to see what we have available and what is permitted.

", "rawMarkdown": "", "votes": null }, { "id": "55441", "postDate": "09/30/2014 20:33:27", "content": "

Excellent! Thank you very much.

\n\n

I understand patient data can't be disclosed (HIPAA and other regulations). But this is certainly of help.

\n\n

AG

", "rawMarkdown": "", "votes": null }, { "id": "55442", "postDate": "09/30/2014 21:04:15", "content": "

Is the gain of the system constant from subject to subject? That is: would the amplification be adjusted if the signal amplitude was unusually small or large?

", "rawMarkdown": "", "votes": null }, { "id": "55444", "postDate": "09/30/2014 21:57:30", "content": "

I have another question.

\n\n

Regarding the montage. It says that the "electrodes are referenced to another electrode outside the head." What does it mean? Is it like a reference montage? In such case, for instance, what you call channel 1 is not really channel 1, it should be channel 1-reference, and then channel 2-reference and so on. Am I right? or Channel 1 is the raw data from channel 1 and I could play changing the references? Or the data is not really "raw" and it is pre referenced?

\n

I think it maybe important for correctly understanding the preictal state, given that if I could play with the referencing point maybe spikes could be viewed more clearly.

\n\n

Thanks.

\n\n

AG

", "rawMarkdown": "", "votes": null }, { "id": "55473", "postDate": "10/01/2014 21:29:17", "content": "

Thanks for the great questions!

\n

No gain adjustment is done for the acquisition. However, do consider that the electrode configuration, length of wires, and other factors may impact the measured voltages (relative to the "true" potentials generated in the brain).

\n

As for the reference question - voltages imply a difference in electrical potentials between two objects or locations. EEG systems (scalp or intracranial) typically have an electrical ground point, and an electrical reference point. (see the wikipedia page on EEG for a more detailed description) The reference point is what is described in the description page. For intracranial EEG a reference electrode is placed near the electrodes in the brain but in a position where it will not record brain activity, muscle artifact, or other activity. For example it might be placed against the skull under the skin flap. However, this location varies depending on the surgical approach and a host of other patient specific factors.

\n

For the human data the reference electrode signal is essentially included in the signal of every channel, i.e. if there were a sine wave oscillation on the reference electrode it would appear inverted in each channel's signal. (although a well-placed reference electrode should only record very slow voltage shifts due to static charge buildup) The data really is "raw" because any measured voltage always has some type of reference.

\n

If you'd like to try different referencing, you have a few options. You can use an average reference, i.e. subtract the mean of all signals from each signal. (As mentioned in the data description this is how the dog data is referenced.) You can also choose a single channel to use as a reference, although bear in mind you'll essentially introduce the inverse of any brain activity recorded on your new reference electrode into all other channels. This may be useful in a few applications, but generally isn't done much clinically. Another possibility (used often in scalp EEG) is to use a Laplacian type of montage, where each electrode is referenced to the average of a set of nearby electrodes. (again see the wikipedia page for more detail) This could remove any electrical activity happening on a larger spatial scale, effectively isolating electrical activity happening directly below a given electrode. Again, I'm not aware of this being done clinically for iEEG, but from a data processing/analytic modeling perspective it could have value.

", "rawMarkdown": "", "votes": null }, { "id": "56156", "postDate": "10/17/2014 05:08:03", "content": "

Hi Ben, you mention we can try different options with the human data regarding the reference signal. Does this mean we are allowed to have an extra preprocessing step for humans that we don't do for the dogs in order to perform this reference signal adjustment? Specifically is this an exception to the following rule?

\n

# Not allowed:
if 'Dog_1' then foo()
if 'Patient_1' then bar()

", "rawMarkdown": "", "votes": null }, { "id": "57045", "postDate": "10/30/2014 14:02:52", "content": "

This is a very good point - in this case though you could apply an average montage to each subject. Re-applying an average montage to the dog data should have essentially no effect.

", "rawMarkdown": "", "votes": null }, { "id": "57091", "postDate": "10/31/2014 05:28:19", "content": "

I thought so too, but I tried applying it to the dogs as well but saw significant decrease in score e.g. 0.77 instead of 0.79.

\n

When you say average montage, do you mean for all the channels, take the mean of sample 0, then subtract that mean from sample 0 of every channel? Then so on for sample 1..N

", "rawMarkdown": "", "votes": null }, { "id": "57435", "postDate": "11/06/2014 21:01:43", "content": "

Average montaging involves subtracting the average of all channels at a particular time point from the signal on each channel at that time point. So the average is calculated across channels, not across time. So I think your description is correct.

", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 53324, "author_name": "franklyn", "author_url": "", "post_date": "09/08/2014 05:57:49", "content": "

I would also like to know the spatial map of the electrodes and the nominal voltage values for each subject (so we can remove spikes)

", "votes": null, "replies": [] }, { "id": 53394, "author_name": "bbrinkm", "author_url": "", "post_date": "09/09/2014 14:11:02", "content": "

This is the same situation as the previous seizure detection challenge (https://www.kaggle.com/c/seizure-detection/forums/t/9354/channel-name-to-location-of-electrode). It really isn't possible to provide an exact electrode location for this contest. For the dogs, the electrodes are positioned horizontally with two 4-contact strips on each side of the brain. Numbering goes from anterior to posterior, starting on the left superior strip. For the humans, there really is no standard system. Electordes are numbered on subdural grids, but grids can be inserted in any orientation. The best one can glean from the naming system is general anatomic coverage.

", "votes": null, "replies": [] }, { "id": 53397, "author_name": "kdoniger", "author_url": "", "post_date": "09/09/2014 15:07:48", "content": "

Just to be clear: this means that channel 1 on one patient could be channel 4 on another patient (or not even exist).

", "votes": null, "replies": [] }, { "id": 53451, "author_name": "bbrinkm", "author_url": "", "post_date": "09/10/2014 16:33:26", "content": "

Yes. That's correct in patients. For the dogs in the contest the arrangement of electrodes is reasonably consistent. For human subjects the placement of electrodes is dictated by the clinical problem and the suspected location of seizure onset, which of course varies from patient to patient.

", "votes": null, "replies": [] }, { "id": 55195, "author_name": "kdoniger", "author_url": "", "post_date": "09/25/2014 00:15:29", "content": "

A related question: On a single subject, are the electrode locations and identifications consistent over all examples?

", "votes": null, "replies": [] }, { "id": 55287, "author_name": "bbrinkm", "author_url": "", "post_date": "09/26/2014 15:10:31", "content": "

Yes. For each subject, each named iEEG channel records activity from a constant location in the brain throughout the entire recording. No electrodes were moved or replaced during the recording, nor were channel names changed.

", "votes": null, "replies": [] }, { "id": 55412, "author_name": "andresgarcia", "author_url": "", "post_date": "09/29/2014 22:17:14", "content": "

Just to clarify.

\n

In the recordings for the dogs, one could sort of assume that channel 1 is frontal, channel 4 is occipital and channels 2 and 3 could be like parieto-occipital?

\n

Also for humans, it could be very useful to know the general localization. Some phenomena are observable in the front cortex with eye movement (like Bell's) or some specific rithms by region, like alfa in the ccipital rebion and beta in the frontal one. If the data from patients is from grids and deep electrodes it could also be helpful to have that kind of data. I know mapping grids is difficult, but knowing if it is temporal or not could also help. Generally when neurosurgeons and epileptologists place grids, they make a small map of it. I wonder if this map is available.

\n\n

Thanks

\n\n

AG

", "votes": null, "replies": [] }, { "id": 55413, "author_name": "andresgarcia", "author_url": "", "post_date": "09/29/2014 22:34:30", "content": "

Also,

\n\n

Going back to the dogs, if there are 2 stripes with 4 electrodes from frontal to occipital numbered 1 to 4 per hemisphere of the brain. Which 2 electrodes are frontal left and which ones are frontal right? The reason for asking this is that probably the electrodes that are close together may see the same electrical activity with different intensity (e.g: the focci is between 2 electrodes).

\n\n

Thanks for the clarification

\n\n

AG

", "votes": null, "replies": [] }, { "id": 55414, "author_name": "mkozine", "author_url": "", "post_date": "09/29/2014 23:05:27", "content": "

AndresGarcia -

\n

Correct me if I am wrong, but I think this info is useless in this particular competition, because the model hast to be electrodes- and dog/patient- invariant, according to the rules. 

", "votes": null, "replies": [] }, { "id": 55415, "author_name": "andresgarcia", "author_url": "", "post_date": "09/29/2014 23:23:13", "content": "

Well, I guess it depends on how are you creating the algorithms. In a sense, one could shoot for a high sensitivity algorithm to include certain activity that looks suspicious of being preictal and then use another algorithm with high specificity to improve your probabilities of correctly finding a spike that will indicate abnormal electrical activity.

\n

I agree, the model has to be dog/patient invariant, but electrodes matter. When reading EEGs (in a usual setting) many times spikes of bipolar behavior are detected. This is that in one electrode the spike looks negative, while it looks positive in the electrode next to it. Hence, the electrical abnormal activity is happening between both electrodes. Sometimes preictal states are missed because misreading such spikes.

\n

Sometimes also, there is a spike in one side of the brain and then electrical activity in the contralateral part, meaning that there is spread (not contiguously but to the contralaeral hemisphere). In order for a good algorith to exists, one could shoot for groing reading by reading by electrode, but also to correlate such readings with other electrodes to increase specificity. I am not sure it is irrelevant...

\n\n

AG

", "votes": null, "replies": [] }, { "id": 55440, "author_name": "bbrinkm", "author_url": "", "post_date": "09/30/2014 20:24:19", "content": "

This may or may not  be helpful, but I'm posting with this email a diagram of the implanted electrodes in the dogs. This applies to all the dogs. As has been noted elsewhere Dog 5 had a channel that went bad, so we omitted it. Bear in mind that these locations are not exact - they represent targeted locations, and there could be quite a bit of variation in exactly where the surgeon placed the electrodes. We do not have post-implant imaging to verify electrode locations.

\n

As for the patient electrode maps and configurations, this is a bit more difficult, and I don't believe I will be able to provide much. Patients do typically have post-implant CT imaging, but I don't believe I can post any of that data here. I may however be able to give a general description of how the electrodes are arranged - I will check on that, both to see what we have available and what is permitted.

", "votes": null, "replies": [] }, { "id": 55441, "author_name": "andresgarcia", "author_url": "", "post_date": "09/30/2014 20:33:27", "content": "

Excellent! Thank you very much.

\n\n

I understand patient data can't be disclosed (HIPAA and other regulations). But this is certainly of help.

\n\n

AG

", "votes": null, "replies": [] }, { "id": 55442, "author_name": "kdoniger", "author_url": "", "post_date": "09/30/2014 21:04:15", "content": "

Is the gain of the system constant from subject to subject? That is: would the amplification be adjusted if the signal amplitude was unusually small or large?

", "votes": null, "replies": [] }, { "id": 55444, "author_name": "andresgarcia", "author_url": "", "post_date": "09/30/2014 21:57:30", "content": "

I have another question.

\n\n

Regarding the montage. It says that the "electrodes are referenced to another electrode outside the head." What does it mean? Is it like a reference montage? In such case, for instance, what you call channel 1 is not really channel 1, it should be channel 1-reference, and then channel 2-reference and so on. Am I right? or Channel 1 is the raw data from channel 1 and I could play changing the references? Or the data is not really "raw" and it is pre referenced?

\n

I think it maybe important for correctly understanding the preictal state, given that if I could play with the referencing point maybe spikes could be viewed more clearly.

\n\n

Thanks.

\n\n

AG

", "votes": null, "replies": [] }, { "id": 55473, "author_name": "bbrinkm", "author_url": "", "post_date": "10/01/2014 21:29:17", "content": "

Thanks for the great questions!

\n

No gain adjustment is done for the acquisition. However, do consider that the electrode configuration, length of wires, and other factors may impact the measured voltages (relative to the "true" potentials generated in the brain).

\n

As for the reference question - voltages imply a difference in electrical potentials between two objects or locations. EEG systems (scalp or intracranial) typically have an electrical ground point, and an electrical reference point. (see the wikipedia page on EEG for a more detailed description) The reference point is what is described in the description page. For intracranial EEG a reference electrode is placed near the electrodes in the brain but in a position where it will not record brain activity, muscle artifact, or other activity. For example it might be placed against the skull under the skin flap. However, this location varies depending on the surgical approach and a host of other patient specific factors.

\n

For the human data the reference electrode signal is essentially included in the signal of every channel, i.e. if there were a sine wave oscillation on the reference electrode it would appear inverted in each channel's signal. (although a well-placed reference electrode should only record very slow voltage shifts due to static charge buildup) The data really is "raw" because any measured voltage always has some type of reference.

\n

If you'd like to try different referencing, you have a few options. You can use an average reference, i.e. subtract the mean of all signals from each signal. (As mentioned in the data description this is how the dog data is referenced.) You can also choose a single channel to use as a reference, although bear in mind you'll essentially introduce the inverse of any brain activity recorded on your new reference electrode into all other channels. This may be useful in a few applications, but generally isn't done much clinically. Another possibility (used often in scalp EEG) is to use a Laplacian type of montage, where each electrode is referenced to the average of a set of nearby electrodes. (again see the wikipedia page for more detail) This could remove any electrical activity happening on a larger spatial scale, effectively isolating electrical activity happening directly below a given electrode. Again, I'm not aware of this being done clinically for iEEG, but from a data processing/analytic modeling perspective it could have value.

", "votes": null, "replies": [] }, { "id": 56156, "author_name": "michaelhills", "author_url": "", "post_date": "10/17/2014 05:08:03", "content": "

Hi Ben, you mention we can try different options with the human data regarding the reference signal. Does this mean we are allowed to have an extra preprocessing step for humans that we don't do for the dogs in order to perform this reference signal adjustment? Specifically is this an exception to the following rule?

\n

# Not allowed:
if 'Dog_1' then foo()
if 'Patient_1' then bar()

", "votes": null, "replies": [] }, { "id": 57045, "author_name": "bbrinkm", "author_url": "", "post_date": "10/30/2014 14:02:52", "content": "

This is a very good point - in this case though you could apply an average montage to each subject. Re-applying an average montage to the dog data should have essentially no effect.

", "votes": null, "replies": [] }, { "id": 57091, "author_name": "michaelhills", "author_url": "", "post_date": "10/31/2014 05:28:19", "content": "

I thought so too, but I tried applying it to the dogs as well but saw significant decrease in score e.g. 0.77 instead of 0.79.

\n

When you say average montage, do you mean for all the channels, take the mean of sample 0, then subtract that mean from sample 0 of every channel? Then so on for sample 1..N

", "votes": null, "replies": [] }, { "id": 57435, "author_name": "bbrinkm", "author_url": "", "post_date": "11/06/2014 21:01:43", "content": "

Average montaging involves subtracting the average of all channels at a particular time point from the signal on each channel at that time point. So the average is calculated across channels, not across time. So I think your description is correct.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "53306": "", "53324": "", "53394": "", "53397": "", "53451": "", "55195": "", "55287": "", "55412": "", "55413": "", "55414": "", "55415": "", "55440": "", "55441": "", "55442": "", "55444": "", "55473": "", "56156": "", "57045": "", "57091": "", "57435": "" }, "source": "meta" }