\n\nThe expert annotators reviewed 50 second long EEG samples plus matched spectrograms covering 10 a minute window centered at the same time and labeled the central 10 seconds.
\n
By far, I understand that the experts annotate according to the EEG and the Spectrogram of four regions on the brain. Notably the spectrogram is the assembled EEG data yet we only have 50 seconds of EEG that is matched to it. Furthermore, spectrogram seems to be a recording of signals from four regions of the brain, yet the EEG data contains data from 19 regions on the brain. Out of curiosity, I wonder how are the impulses converted between them.
Also, why is the label a label on the central 10 seconds. Shouldn't it be a label on both the 50 seconds EEG and 10 mins spectrogram?
\nI sincerely appreciate any response that addresses these concerns. Happy Kaggling!😉
", "messages": [ { "id": "2610351", "postDate": "01/20/2024 05:13:18", "content": "\n\nThe expert annotators reviewed 50 second long EEG samples plus matched spectrograms covering 10 a minute window centered at the same time and labeled the central 10 seconds.
\n
By far, I understand that the experts annotate according to the EEG and the Spectrogram of four regions on the brain. Notably the spectrogram is the assembled EEG data yet we only have 50 seconds of EEG that is matched to it. Furthermore, spectrogram seems to be a recording of signals from four regions of the brain, yet the EEG data contains data from 19 regions on the brain. Out of curiosity, I wonder how are the impulses converted between them.
Also, why is the label a label on the central 10 seconds. Shouldn't it be a label on both the 50 seconds EEG and 10 mins spectrogram?
\nI sincerely appreciate any response that addresses these concerns. Happy Kaggling!😉
", "rawMarkdown": ">The expert annotators reviewed 50 second long EEG samples plus matched spectrograms covering 10 a minute window centered at the same time and labeled the central 10 seconds.\n\nBy far, I understand that the experts annotate according to the EEG and the Spectrogram of four regions on the brain. Notably the spectrogram is the `assembled EEG data` yet we only have 50 seconds of EEG that is matched to it. Furthermore, spectrogram seems to be a recording of signals from four regions of the brain, yet the EEG data contains data from 19 regions on the brain. Out of curiosity, I wonder how are the impulses converted between them.\n\nAlso, why is the label a label on the central 10 seconds. Shouldn't it be a label on both the 50 seconds EEG and 10 mins spectrogram?\n\n---\n\nI sincerely appreciate any response that addresses these concerns. Happy Kaggling!😉", "votes": null }, { "id": "2610364", "postDate": "01/20/2024 05:22:47", "content": "@renyiwei Also, why is the label a label on the central 10 seconds. Shouldn't it be a label on both the 50 seconds EEG and 10 mins spectrogram?
\n\n", "rawMarkdown": "renyiwei Also, why is the label a label on the central 10 seconds. Shouldn't it be a label on both the 50 seconds EEG and 10 mins spectrogram?\n> we have many egg_id for each patient_id\n> for each egg_id, we have egg_sub_id starts with offset eeg_label_offset_seconds\n> while label the egg_sub_id, expert decided only based on middle 10secs out of 50secs\ncheck this discussion - **Understanding Competition Data and EfficientNetB2 Starter - LB 0.57**", "votes": null }, { "id": "2610431", "postDate": "01/20/2024 06:05:26", "content": "we have many egg_id for each patient_id
\n
\n for each egg_id, we have egg_sub_id starts with offset eeg_label_offset_seconds
\n while label the egg_sub_id, expert decided only based on middle 10secs out of 50secs
\n check this discussion - Understanding Competition Data and EfficientNetB2 Starter - LB 0.57
Hi @seshurajup. Correct me if I am wrong. I think what they mean is that a very short period of impulse is categorized into multiple classes, (i.e. LPD, seizure, etc). Yet experts need additional information to better classify that 10 seconds of impulses. Likewise, 50s of EEG and 10 mins of spectrogram is provided to us
", "rawMarkdown": "Hi @seshurajup. Correct me if I am wrong. I think what they mean is that a very short period of impulse is categorized into multiple classes, (i.e. LPD, seizure, etc). Yet experts need additional information to better classify that 10 seconds of impulses. Likewise, 50s of EEG and 10 mins of spectrogram is provided to us", "votes": null }, { "id": "2610434", "postDate": "01/20/2024 06:07:56", "content": "@renyiwei even expert categorised based on 10secs, he observe 50secs sequence and 4 LL, LP, RP, RR spectrograms => then voted is my understanding.
", "rawMarkdown": "renyiwei even expert categorised based on 10secs, he observe 50secs sequence and 4 LL, LP, RP, RR spectrograms => then voted is my understanding.", "votes": null }, { "id": "2661494", "postDate": "02/21/2024 09:49:33", "content": "Bit late but I hope this answers your question
\nSee comment here. https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/467105
\nEssentially, too expensive to give full EEG recordings
EEG is sampled at 200 Hz whereas Spectrogram is sampled at 0.5 Hz.
\nHence if we wanted the corresponding 10 minutes of EEG that would be 10 * 60 * 200 = 120k
\nBut realistically experts only look at the central 50 seconds
Note a spectrogram is a function of the EEG. (I.e. take 2 second chunks and compute the Fourier Transform)
\nThe spectrograms that we are given correspond to 4 regions of the brain. These are a functions of the spectrogram for individual EEG Time Series. See here for more details https://www.kaggle.com/code/cdeotte/how-to-make-spectrogram-from-eeg
\nHaving spectrogram corresponding to regions is more interpretable as our classes want to distinguish between Lateral (happening in only one of the regions) and Global (happening in all of the regions).
\nIf you want to see more details on how they're constructed see the paper pointed to here
\nhttps://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/477682
Ideally we would like to be able to classify at every point in time. So asking for what happens in a region of 10 seconds is an approximation to that.
\nIndeed, we are given more data than this. I.e. 50 seconds of EEG and 10 minutes of spectrogram. However, this should act like context and give more information. Though it is up to you if you want to ignore it and only focus on the central 10 seconds. This is a modelling choice.
", "rawMarkdown": "Bit late but I hope this answers your question\n\n# Why we only have 50 seconds matched to 10 minutes of Spectrogram\nSee comment here. https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/467105\nEssentially, too expensive to give full EEG recordings\n\nEEG is sampled at 200 Hz whereas Spectrogram is sampled at 0.5 Hz.\nHence if we wanted the corresponding 10 minutes of EEG that would be 10 * 60 * 200 = 120k\nBut realistically experts only look at the central 50 seconds\n\nNote a spectrogram is a function of the EEG. (I.e. take 2 second chunks and compute the Fourier Transform)\n\n# Why there are 19 EEG time series yet only 4 spectrograms\nThe spectrograms that we are given correspond to 4 regions of the brain. These are a functions of the spectrogram for individual EEG Time Series. See here for more details https://www.kaggle.com/code/cdeotte/how-to-make-spectrogram-from-eeg\n\nHaving spectrogram corresponding to regions is more interpretable as our classes want to distinguish between Lateral (happening in only one of the regions) and Global (happening in all of the regions).\n\nIf you want to see more details on how they're constructed see the paper pointed to here\nhttps://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/477682\n\n# Why are we labelling the 10 central seconds\nIdeally we would like to be able to classify at every point in time. So asking for what happens in a region of 10 seconds is an approximation to that.\n\nIndeed, we are given more data than this. I.e. 50 seconds of EEG and 10 minutes of spectrogram. However, this should act like context and give more information. Though it is up to you if you want to ignore it and only focus on the central 10 seconds. This is a modelling choice.", "votes": null } ], "comments": [ { "id": 2610364, "author_name": "seshurajup", "author_url": "", "post_date": "01/20/2024 05:22:47", "content": "@renyiwei Also, why is the label a label on the central 10 seconds. Shouldn't it be a label on both the 50 seconds EEG and 10 mins spectrogram?
\n\n", "votes": null, "replies": [ { "id": 2610431, "author_name": "renyiwei", "author_url": "", "post_date": "01/20/2024 06:05:26", "content": "we have many egg_id for each patient_id
\n
\n for each egg_id, we have egg_sub_id starts with offset eeg_label_offset_seconds
\n while label the egg_sub_id, expert decided only based on middle 10secs out of 50secs
\n check this discussion - Understanding Competition Data and EfficientNetB2 Starter - LB 0.57
Hi @seshurajup. Correct me if I am wrong. I think what they mean is that a very short period of impulse is categorized into multiple classes, (i.e. LPD, seizure, etc). Yet experts need additional information to better classify that 10 seconds of impulses. Likewise, 50s of EEG and 10 mins of spectrogram is provided to us
", "votes": null, "replies": [ { "id": 2610434, "author_name": "seshurajup", "author_url": "", "post_date": "01/20/2024 06:07:56", "content": "@renyiwei even expert categorised based on 10secs, he observe 50secs sequence and 4 LL, LP, RP, RR spectrograms => then voted is my understanding.
", "votes": null, "replies": [] } ] } ] }, { "id": 2661494, "author_name": "delacruzdanilojr", "author_url": "", "post_date": "02/21/2024 09:49:33", "content": "Bit late but I hope this answers your question
\nSee comment here. https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/467105
\nEssentially, too expensive to give full EEG recordings
EEG is sampled at 200 Hz whereas Spectrogram is sampled at 0.5 Hz.
\nHence if we wanted the corresponding 10 minutes of EEG that would be 10 * 60 * 200 = 120k
\nBut realistically experts only look at the central 50 seconds
Note a spectrogram is a function of the EEG. (I.e. take 2 second chunks and compute the Fourier Transform)
\nThe spectrograms that we are given correspond to 4 regions of the brain. These are a functions of the spectrogram for individual EEG Time Series. See here for more details https://www.kaggle.com/code/cdeotte/how-to-make-spectrogram-from-eeg
\nHaving spectrogram corresponding to regions is more interpretable as our classes want to distinguish between Lateral (happening in only one of the regions) and Global (happening in all of the regions).
\nIf you want to see more details on how they're constructed see the paper pointed to here
\nhttps://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/477682
Ideally we would like to be able to classify at every point in time. So asking for what happens in a region of 10 seconds is an approximation to that.
\nIndeed, we are given more data than this. I.e. 50 seconds of EEG and 10 minutes of spectrogram. However, this should act like context and give more information. Though it is up to you if you want to ignore it and only focus on the central 10 seconds. This is a modelling choice.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2610351": ">The expert annotators reviewed 50 second long EEG samples plus matched spectrograms covering 10 a minute window centered at the same time and labeled the central 10 seconds.\n\nBy far, I understand that the experts annotate according to the EEG and the Spectrogram of four regions on the brain. Notably the spectrogram is the `assembled EEG data` yet we only have 50 seconds of EEG that is matched to it. Furthermore, spectrogram seems to be a recording of signals from four regions of the brain, yet the EEG data contains data from 19 regions on the brain. Out of curiosity, I wonder how are the impulses converted between them.\n\nAlso, why is the label a label on the central 10 seconds. Shouldn't it be a label on both the 50 seconds EEG and 10 mins spectrogram?\n\n---\n\nI sincerely appreciate any response that addresses these concerns. Happy Kaggling!😉", "2610364": "renyiwei Also, why is the label a label on the central 10 seconds. Shouldn't it be a label on both the 50 seconds EEG and 10 mins spectrogram?\n> we have many egg_id for each patient_id\n> for each egg_id, we have egg_sub_id starts with offset eeg_label_offset_seconds\n> while label the egg_sub_id, expert decided only based on middle 10secs out of 50secs\ncheck this discussion - **Understanding Competition Data and EfficientNetB2 Starter - LB 0.57**", "2610431": "Hi @seshurajup. Correct me if I am wrong. I think what they mean is that a very short period of impulse is categorized into multiple classes, (i.e. LPD, seizure, etc). Yet experts need additional information to better classify that 10 seconds of impulses. Likewise, 50s of EEG and 10 mins of spectrogram is provided to us", "2610434": "renyiwei even expert categorised based on 10secs, he observe 50secs sequence and 4 LL, LP, RP, RR spectrograms => then voted is my understanding.", "2661494": "Bit late but I hope this answers your question\n\n# Why we only have 50 seconds matched to 10 minutes of Spectrogram\nSee comment here. https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/467105\nEssentially, too expensive to give full EEG recordings\n\nEEG is sampled at 200 Hz whereas Spectrogram is sampled at 0.5 Hz.\nHence if we wanted the corresponding 10 minutes of EEG that would be 10 * 60 * 200 = 120k\nBut realistically experts only look at the central 50 seconds\n\nNote a spectrogram is a function of the EEG. (I.e. take 2 second chunks and compute the Fourier Transform)\n\n# Why there are 19 EEG time series yet only 4 spectrograms\nThe spectrograms that we are given correspond to 4 regions of the brain. These are a functions of the spectrogram for individual EEG Time Series. See here for more details https://www.kaggle.com/code/cdeotte/how-to-make-spectrogram-from-eeg\n\nHaving spectrogram corresponding to regions is more interpretable as our classes want to distinguish between Lateral (happening in only one of the regions) and Global (happening in all of the regions).\n\nIf you want to see more details on how they're constructed see the paper pointed to here\nhttps://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/477682\n\n# Why are we labelling the 10 central seconds\nIdeally we would like to be able to classify at every point in time. So asking for what happens in a region of 10 seconds is an approximation to that.\n\nIndeed, we are given more data than this. I.e. 50 seconds of EEG and 10 minutes of spectrogram. However, this should act like context and give more information. Though it is up to you if you want to ignore it and only focus on the central 10 seconds. This is a modelling choice." }, "source": "meta" }