Learning some domain knowledge the hard way. Girlfriend had brain bleeds, needed surgery at Christmas to stop the bleeds and was in the process of recovering. Recently she started having seizures. She's scheduled for an EEG in a week.
\nWhat I have learned so far would seem to suggest that the patients_id has some valuable information to extract, perhaps easily usable for those using tabular methods. Of course, like any good kaggle competion we know nothing about the patients. We do know the number of patients and the number of rows of data per patient.
\nMy girlfriend was proscribed a drug for seizure prevention. They had hoped to get her down to a minimum amount per day before the eeg. In her case, that's not going to happen as seizures returned on the lower dose, and she's likely to end up with high dose after a consult tomorrow.
\nCan we expect to see medication effects with variation, over the count of data rows per patient?
For the test she's to arrive with clean hair, with no hair products. Could we expect to see a difference per patients like her (full head of long great looking hair) vs scans for me (pretty much bald in all probe areas).
\nFor the test - no caffeine for 12 hours prior. Around the 12 hour mark is the point were I start to get a lack of caffeine headache. Can we expect differences per patient based on caffeine or other similar products.
\nFor the test - suggestion to stay up late the night before the test and try to arrive in a tired state, rather than well rested state. According to ChatGPT the normal recommendation is to arrive well rested. Different seizure types can have different directions for this feature.
\nHopefully a second eeg in a few months will show a huge difference in seizure related signals. Can we expect that some patient scans show the last detectable seizure signal close to full recovery?
", "messages": [ { "id": "2625753", "postDate": "01/29/2024 15:11:12", "content": "Learning some domain knowledge the hard way. Girlfriend had brain bleeds, needed surgery at Christmas to stop the bleeds and was in the process of recovering. Recently she started having seizures. She's scheduled for an EEG in a week.
\nWhat I have learned so far would seem to suggest that the patients_id has some valuable information to extract, perhaps easily usable for those using tabular methods. Of course, like any good kaggle competion we know nothing about the patients. We do know the number of patients and the number of rows of data per patient.
\nMy girlfriend was proscribed a drug for seizure prevention. They had hoped to get her down to a minimum amount per day before the eeg. In her case, that's not going to happen as seizures returned on the lower dose, and she's likely to end up with high dose after a consult tomorrow.
\nCan we expect to see medication effects with variation, over the count of data rows per patient?
For the test she's to arrive with clean hair, with no hair products. Could we expect to see a difference per patients like her (full head of long great looking hair) vs scans for me (pretty much bald in all probe areas).
\nFor the test - no caffeine for 12 hours prior. Around the 12 hour mark is the point were I start to get a lack of caffeine headache. Can we expect differences per patient based on caffeine or other similar products.
\nFor the test - suggestion to stay up late the night before the test and try to arrive in a tired state, rather than well rested state. According to ChatGPT the normal recommendation is to arrive well rested. Different seizure types can have different directions for this feature.
\nHopefully a second eeg in a few months will show a huge difference in seizure related signals. Can we expect that some patient scans show the last detectable seizure signal close to full recovery?
", "rawMarkdown": "Learning some domain knowledge the hard way. Girlfriend had brain bleeds, needed surgery at Christmas to stop the bleeds and was in the process of recovering. Recently she started having seizures. She's scheduled for an EEG in a week.\n\nWhat I have learned so far would seem to suggest that the patients_id has some valuable information to extract, perhaps easily usable for those using tabular methods. Of course, like any good kaggle competion we know nothing about the patients. We do know the number of patients and the number of rows of data per patient.\n\nMy girlfriend was proscribed a drug for seizure prevention. They had hoped to get her down to a minimum amount per day before the eeg. In her case, that's not going to happen as seizures returned on the lower dose, and she's likely to end up with high dose after a consult tomorrow. \nCan we expect to see medication effects with variation, over the count of data rows per patient?\n\nFor the test she's to arrive with clean hair, with no hair products. Could we expect to see a difference per patients like her (full head of long great looking hair) vs scans for me (pretty much bald in all probe areas).\n\nFor the test - no caffeine for 12 hours prior. Around the 12 hour mark is the point were I start to get a lack of caffeine headache. Can we expect differences per patient based on caffeine or other similar products.\n\nFor the test - suggestion to stay up late the night before the test and try to arrive in a tired state, rather than well rested state. According to ChatGPT the normal recommendation is to arrive well rested. Different seizure types can have different directions for this feature.\n\nHopefully a second eeg in a few months will show a huge difference in seizure related signals. Can we expect that some patient scans show the last detectable seizure signal close to full recovery?", "votes": null }, { "id": "2625766", "postDate": "01/29/2024 15:20:06", "content": "Best wishes to your girlfriend, hope your girlfriend gets well soon.
", "rawMarkdown": "Best wishes to your girlfriend, hope your girlfriend gets well soon.", "votes": null }, { "id": "2625796", "postDate": "01/29/2024 16:04:43", "content": "hope your girlfriend gets well soon bro, and may all the efforts put in by the kagglers in every game actually apply and help people
", "rawMarkdown": "hope your girlfriend gets well soon bro, and may all the efforts put in by the kagglers in every game actually apply and help people", "votes": null }, { "id": "2625872", "postDate": "01/29/2024 17:19:04", "content": "I think this small variations ( hair, caffeine, rest…) are already covered by agreement intervals. You can create a feature and calculate the percentual of votes that each brain activity class received. High agreements are expected to be correlated with exams with clear signals.
\nOther important thing to mention is that diagnosis with more expert votes are more reliable than diagnosis with fewer expert votes.
Thanks - it's amazing how much life can change with one small incident. Her dog chased after a rabbit on walk, while on a retractable leash. See did not see him take off and got pulled to the ground with he hit the end of the leash.
", "rawMarkdown": "Thanks - it's amazing how much life can change with one small incident. Her dog chased after a rabbit on walk, while on a retractable leash. See did not see him take off and got pulled to the ground with he hit the end of the leash.", "votes": null }, { "id": "2625952", "postDate": "01/29/2024 17:51:27", "content": "Thanks - does make a person appreciate the value of what can be gained by kagglers helping to improve medical diagnosis.
", "rawMarkdown": "Thanks - does make a person appreciate the value of what can be gained by kagglers helping to improve medical diagnosis.", "votes": null }, { "id": "2625954", "postDate": "01/29/2024 17:53:36", "content": "It's sorry to hear that. Life is so uncertain, just like the shake in private leaderboard.
", "rawMarkdown": "It's sorry to hear that. Life is so uncertain, just like the shake in private leaderboard.", "votes": null }, { "id": "2625959", "postDate": "01/29/2024 17:56:40", "content": "Not sure I have seen any shared notebooks were 'within patient' features have been generated that address this - I need to look again.
", "rawMarkdown": "Not sure I have seen any shared notebooks were 'within patient' features have been generated that address this - I need to look again.", "votes": null }, { "id": "2625960", "postDate": "01/29/2024 17:57:36", "content": "Now that's funny !!!
", "rawMarkdown": "Now that's funny !!!", "votes": null }, { "id": "2626548", "postDate": "01/30/2024 05:26:56", "content": "You are right, and I am trying.
", "rawMarkdown": "You are right, and I am trying.", "votes": null }, { "id": "2627532", "postDate": "01/30/2024 18:06:03", "content": "I've been analyzing the data structure of the competition and a question arose:
\n1- Are the patient_ids in the test dataset the same as those in the training dataset?
\n If yes, this could open up a range of possibilities for exploring individual patient characteristics in the modeling phase.
\n On the other hand, if the patient_ids in the test dataset are different from those in the training dataset, it suggests that a 'group fold' type of separation method was applied in data preparation.
Does not appear that any train patients are in test.
\nThere are two types of possible ‘within patient’ information. Multiple target values within a single scan, and multiple eeg ids for the same patient. I think both are still source of information for models even if no train patients are in test.
\n20 years ago I would have called the first measurement error and the second repeatability.
", "rawMarkdown": "Does not appear that any train patients are in test. \n\nThere are two types of possible ‘within patient’ information. Multiple target values within a single scan, and multiple eeg ids for the same patient. I think both are still source of information for models even if no train patients are in test. \n\n20 years ago I would have called the first measurement error and the second repeatability.", "votes": null }, { "id": "2629574", "postDate": "01/31/2024 21:36:43", "content": "@PC JIMMY All the best for the lady. I had a similar incident some years ago. Except good doctors, methods and diagnoses a good amount of luck is needed
", "rawMarkdown": "PC JIMMY All the best for the lady. I had a similar incident some years ago. Except good doctors, methods and diagnoses a good amount of luck is needed", "votes": null }, { "id": "2668085", "postDate": "02/25/2024 14:00:50", "content": "Girlfriend update:
\nHad a CT Scan a few days ago. Report shows huge reduction of the amount of blood on the brain. No seizure's since I first posted this a month ago.
\nAfter getting out of hospital in December she could barely walk. She's now doing 6 miles or more per day, dragging me along for 3 of them in the afternoon.
\nStill on meds, but expecting a reduction in the amount at this coming weeks doctor visit.
\nI am guessing based on her improvement that there will be no second eeg. It was an hour long session in hospital so I assume a bit costly.
\nTake-aways from this for our competition.
\nBest wishes to your girlfriend, hope your girlfriend gets well soon.
", "votes": null, "replies": [ { "id": 2625947, "author_name": "pcjimmmy", "author_url": "", "post_date": "01/29/2024 17:49:32", "content": "Thanks - it's amazing how much life can change with one small incident. Her dog chased after a rabbit on walk, while on a retractable leash. See did not see him take off and got pulled to the ground with he hit the end of the leash.
", "votes": null, "replies": [ { "id": 2625954, "author_name": "tonyarobertson", "author_url": "", "post_date": "01/29/2024 17:53:36", "content": "It's sorry to hear that. Life is so uncertain, just like the shake in private leaderboard.
", "votes": null, "replies": [ { "id": 2625960, "author_name": "pcjimmmy", "author_url": "", "post_date": "01/29/2024 17:57:36", "content": "Now that's funny !!!
", "votes": null, "replies": [] } ] } ] } ] }, { "id": 2625796, "author_name": "gentlezdh", "author_url": "", "post_date": "01/29/2024 16:04:43", "content": "hope your girlfriend gets well soon bro, and may all the efforts put in by the kagglers in every game actually apply and help people
", "votes": null, "replies": [ { "id": 2625952, "author_name": "pcjimmmy", "author_url": "", "post_date": "01/29/2024 17:51:27", "content": "Thanks - does make a person appreciate the value of what can be gained by kagglers helping to improve medical diagnosis.
", "votes": null, "replies": [ { "id": 2629574, "author_name": "georgem", "author_url": "", "post_date": "01/31/2024 21:36:43", "content": "@PC JIMMY All the best for the lady. I had a similar incident some years ago. Except good doctors, methods and diagnoses a good amount of luck is needed
", "votes": null, "replies": [] } ] } ] }, { "id": 2625872, "author_name": "serjhenrique", "author_url": "", "post_date": "01/29/2024 17:19:04", "content": "I think this small variations ( hair, caffeine, rest…) are already covered by agreement intervals. You can create a feature and calculate the percentual of votes that each brain activity class received. High agreements are expected to be correlated with exams with clear signals.
\nOther important thing to mention is that diagnosis with more expert votes are more reliable than diagnosis with fewer expert votes.
Not sure I have seen any shared notebooks were 'within patient' features have been generated that address this - I need to look again.
", "votes": null, "replies": [] } ] }, { "id": 2626548, "author_name": "sunyuri", "author_url": "", "post_date": "01/30/2024 05:26:56", "content": "You are right, and I am trying.
", "votes": null, "replies": [] }, { "id": 2627532, "author_name": "rafaelzimmermann1", "author_url": "", "post_date": "01/30/2024 18:06:03", "content": "I've been analyzing the data structure of the competition and a question arose:
\n1- Are the patient_ids in the test dataset the same as those in the training dataset?
\n If yes, this could open up a range of possibilities for exploring individual patient characteristics in the modeling phase.
\n On the other hand, if the patient_ids in the test dataset are different from those in the training dataset, it suggests that a 'group fold' type of separation method was applied in data preparation.
Does not appear that any train patients are in test.
\nThere are two types of possible ‘within patient’ information. Multiple target values within a single scan, and multiple eeg ids for the same patient. I think both are still source of information for models even if no train patients are in test.
\n20 years ago I would have called the first measurement error and the second repeatability.
", "votes": null, "replies": [] } ] }, { "id": 2668085, "author_name": "pcjimmmy", "author_url": "", "post_date": "02/25/2024 14:00:50", "content": "Girlfriend update:
\nHad a CT Scan a few days ago. Report shows huge reduction of the amount of blood on the brain. No seizure's since I first posted this a month ago.
\nAfter getting out of hospital in December she could barely walk. She's now doing 6 miles or more per day, dragging me along for 3 of them in the afternoon.
\nStill on meds, but expecting a reduction in the amount at this coming weeks doctor visit.
\nI am guessing based on her improvement that there will be no second eeg. It was an hour long session in hospital so I assume a bit costly.
\nTake-aways from this for our competition.
\n