{ "id": 9917, "title": "Errors in data?", "url": "/competitions/seizure-detection/discussion/9917", "author_name": "", "post_date": "2014-08-01T11:40:13.193Z", "votes": 2, "comment_count": 3, "views": 922, "content": "

Hey,

\n

I did a small exploration of Patient 4 and I noticed some really strange stuff. Right-click an image and select "view image" for a much better detail.

\n

\n

Each blue line is a time series from an electrode. These plots look very suspicious, e.g. the lower left plot contains 6 straight lines and the rest are perfect sinusoids. They are not the only occurrences, there are about 10-20 such in the data of Patient 4, both in train and test sets.

\n

Is it intended? What about other patients?

\n

As a comparison, below are plots representing what I think correct time series are:

\n

", "messages": [ { "id": "51473", "postDate": "08/01/2014 11:40:13", "content": "

Hey,

\n

I did a small exploration of Patient 4 and I noticed some really strange stuff. Right-click an image and select "view image" for a much better detail.

\n

\n

Each blue line is a time series from an electrode. These plots look very suspicious, e.g. the lower left plot contains 6 straight lines and the rest are perfect sinusoids. They are not the only occurrences, there are about 10-20 such in the data of Patient 4, both in train and test sets.

\n

Is it intended? What about other patients?

\n

As a comparison, below are plots representing what I think correct time series are:

\n

", "rawMarkdown": "", "votes": null }, { "id": "51476", "postDate": "08/01/2014 12:02:03", "content": "

This is correct, there are no errors in the data.

\n

The strange stuff you see could be an epileptic seizure or perhaps even some measuring artifacts. Consider it part of the problem and try to get a good score.

", "rawMarkdown": "", "votes": null }, { "id": "51479", "postDate": "08/01/2014 12:58:17", "content": "

[quote=Triskelion;51476]

\n

This is correct, there are no errors in the data.

\n

The strange stuff you see could be an epileptic seizure or perhaps even some measuring artifacts. Consider it part of the problem and try to get a good score.

\n

[/quote]

\n

I think these are indeed measurement errors. Of course the organizer wants to have a very robust system. I just wanted to raise the possible issue of presence of such errors and make other Kagglers aware of them.

", "rawMarkdown": "", "votes": null }, { "id": "51481", "postDate": "08/01/2014 14:11:18", "content": "

These are indeed acquisition artifacts in the iEEG. The perfect sinusoids Blaine found are from electromagnetic induction - the recording equipment is very sensitive and picks up 60 Hz EM from electric lights, transformers, televisions, and other devices. We shield the system carefully but occasionally some shielding will come loose, or a technician may adjust a connection.

\n

The first artifact you show is probably patient motion. It's pretty common in these recordings. The second artifact is less familiar, but most likely a slow static depolarization in the recording system's amplifiers.

\n

These artifacts are inevitable in intracranial EEG, and we wanted them to be part of the challenge. In particular for the motion artifact - you can imagine it would be sub-optimal for neurostimulation to occur every time a patient moves.

\n

Thanks for the great questions!

", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 51476, "author_name": "triskelion", "author_url": "", "post_date": "08/01/2014 12:02:03", "content": "

This is correct, there are no errors in the data.

\n

The strange stuff you see could be an epileptic seizure or perhaps even some measuring artifacts. Consider it part of the problem and try to get a good score.

", "votes": null, "replies": [] }, { "id": 51479, "author_name": "rafalcycon", "author_url": "", "post_date": "08/01/2014 12:58:17", "content": "

[quote=Triskelion;51476]

\n

This is correct, there are no errors in the data.

\n

The strange stuff you see could be an epileptic seizure or perhaps even some measuring artifacts. Consider it part of the problem and try to get a good score.

\n

[/quote]

\n

I think these are indeed measurement errors. Of course the organizer wants to have a very robust system. I just wanted to raise the possible issue of presence of such errors and make other Kagglers aware of them.

", "votes": null, "replies": [] }, { "id": 51481, "author_name": "bbrinkm", "author_url": "", "post_date": "08/01/2014 14:11:18", "content": "

These are indeed acquisition artifacts in the iEEG. The perfect sinusoids Blaine found are from electromagnetic induction - the recording equipment is very sensitive and picks up 60 Hz EM from electric lights, transformers, televisions, and other devices. We shield the system carefully but occasionally some shielding will come loose, or a technician may adjust a connection.

\n

The first artifact you show is probably patient motion. It's pretty common in these recordings. The second artifact is less familiar, but most likely a slow static depolarization in the recording system's amplifiers.

\n

These artifacts are inevitable in intracranial EEG, and we wanted them to be part of the challenge. In particular for the motion artifact - you can imagine it would be sub-optimal for neurostimulation to occur every time a patient moves.

\n

Thanks for the great questions!

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "51473": "", "51476": "", "51479": "", "51481": "" }, "source": "meta" }