Cole Korponay
@ckorponay.bsky.social
🧠 Instructor in Psychiatry
@ Harvard Medical School | McLean Hospital
| Trying to squeeze better brain juice from fMRI
@ Harvard Medical School | McLean Hospital
| Trying to squeeze better brain juice from fMRI
And yes, sLFO changes in response to drug administration appear to be most directly driven by changes in cerebral blood flow that stem from arousal-related physiological changes (e.g., to breathing and heart rates) attributable to the drug.
August 27, 2025 at 2:12 PM
And yes, sLFO changes in response to drug administration appear to be most directly driven by changes in cerebral blood flow that stem from arousal-related physiological changes (e.g., to breathing and heart rates) attributable to the drug.
Great questions! The sLFO is a non-neuronal component of the global signal (which includes both neuronal and non-neuronal components). If you're interested in getting the cleanest read-out of neuronal signals possible, then regressing out the sLFO is always going to help with that.
August 27, 2025 at 2:08 PM
Great questions! The sLFO is a non-neuronal component of the global signal (which includes both neuronal and non-neuronal components). If you're interested in getting the cleanest read-out of neuronal signals possible, then regressing out the sLFO is always going to help with that.
What's the take-away?
Measurements of non-cognitive mental health and measurements of brain function during common fMRI tasks have little to do with one another
Akin to cardiac stress tests, more affectively provocative tasks may be needed to evoke🧠activity w/ MH relevance
Measurements of non-cognitive mental health and measurements of brain function during common fMRI tasks have little to do with one another
Akin to cardiac stress tests, more affectively provocative tasks may be needed to evoke🧠activity w/ MH relevance
July 10, 2025 at 7:05 PM
What's the take-away?
Measurements of non-cognitive mental health and measurements of brain function during common fMRI tasks have little to do with one another
Akin to cardiac stress tests, more affectively provocative tasks may be needed to evoke🧠activity w/ MH relevance
Measurements of non-cognitive mental health and measurements of brain function during common fMRI tasks have little to do with one another
Akin to cardiac stress tests, more affectively provocative tasks may be needed to evoke🧠activity w/ MH relevance
We also found that:
1) Brain data and mental health data sorted subjects into discordant sets of subtypes
2) Subjects' relatively similar evoked brain activity profiles failed to reflect the comparatively diverse set of mental health profiles across subjects
1) Brain data and mental health data sorted subjects into discordant sets of subtypes
2) Subjects' relatively similar evoked brain activity profiles failed to reflect the comparatively diverse set of mental health profiles across subjects
July 10, 2025 at 7:05 PM
We also found that:
1) Brain data and mental health data sorted subjects into discordant sets of subtypes
2) Subjects' relatively similar evoked brain activity profiles failed to reflect the comparatively diverse set of mental health profiles across subjects
1) Brain data and mental health data sorted subjects into discordant sets of subtypes
2) Subjects' relatively similar evoked brain activity profiles failed to reflect the comparatively diverse set of mental health profiles across subjects
Multi-network based prediction was also only successful for cognition and proc speed.
Importantly, the composite scores used for each MH domain all had high measurement reliability and similar interindividual variability.
Importantly, the composite scores used for each MH domain all had high measurement reliability and similar interindividual variability.
July 10, 2025 at 7:05 PM
Multi-network based prediction was also only successful for cognition and proc speed.
Importantly, the composite scores used for each MH domain all had high measurement reliability and similar interindividual variability.
Importantly, the composite scores used for each MH domain all had high measurement reliability and similar interindividual variability.
Across all 77 networks, none robustly or reproducibly encoded #WellBeing, #Internalizing or #SubstanceUse, even during tasks w/ socioemotional content.
Instead,🧠network activity near-exclusively encoded individual differences in #cognition across all paradigms.
Instead,🧠network activity near-exclusively encoded individual differences in #cognition across all paradigms.
July 10, 2025 at 7:05 PM
Across all 77 networks, none robustly or reproducibly encoded #WellBeing, #Internalizing or #SubstanceUse, even during tasks w/ socioemotional content.
Instead,🧠network activity near-exclusively encoded individual differences in #cognition across all paradigms.
Instead,🧠network activity near-exclusively encoded individual differences in #cognition across all paradigms.
In our new paper, we searched for markers of normative MH variability in all the evoked🧠networks we could find across a slew of fMRI paradigms w/ putative MH relevance
(We've released maps of all 77 networks in a new "Task-Evoked Network Atlas"): identifiers.org/neurovault.c...
(We've released maps of all 77 networks in a new "Task-Evoked Network Atlas"): identifiers.org/neurovault.c...
July 10, 2025 at 7:05 PM
In our new paper, we searched for markers of normative MH variability in all the evoked🧠networks we could find across a slew of fMRI paradigms w/ putative MH relevance
(We've released maps of all 77 networks in a new "Task-Evoked Network Atlas"): identifiers.org/neurovault.c...
(We've released maps of all 77 networks in a new "Task-Evoked Network Atlas"): identifiers.org/neurovault.c...
Multi-network based prediction was also only successful for cognition and proc speed.
Importantly, the composite scores used for each MH domain all had high measurement reliability and similar interindividual variability.
Importantly, the composite scores used for each MH domain all had high measurement reliability and similar interindividual variability.
July 10, 2025 at 5:32 PM
Multi-network based prediction was also only successful for cognition and proc speed.
Importantly, the composite scores used for each MH domain all had high measurement reliability and similar interindividual variability.
Importantly, the composite scores used for each MH domain all had high measurement reliability and similar interindividual variability.
Across all 77 networks, none robustly or reproducibly encoded #WellBeing, #Internalizing or #SubstanceUse, even during tasks w/ socioemotional content.
Instead,🧠network activity near-exclusively encoded individual differences in #cognition across all paradigms.
Instead,🧠network activity near-exclusively encoded individual differences in #cognition across all paradigms.
July 10, 2025 at 5:32 PM
Across all 77 networks, none robustly or reproducibly encoded #WellBeing, #Internalizing or #SubstanceUse, even during tasks w/ socioemotional content.
Instead,🧠network activity near-exclusively encoded individual differences in #cognition across all paradigms.
Instead,🧠network activity near-exclusively encoded individual differences in #cognition across all paradigms.
In our new paper, we searched for markers of normative MH variability in all the evoked🧠networks we could find across a slew of fMRI paradigms w/ putative MH relevance
(We've released maps of all 77 networks in a new "Task-Evoked Network Atlas"): identifiers.org/neurovault.c...
(We've released maps of all 77 networks in a new "Task-Evoked Network Atlas"): identifiers.org/neurovault.c...
July 10, 2025 at 5:32 PM
In our new paper, we searched for markers of normative MH variability in all the evoked🧠networks we could find across a slew of fMRI paradigms w/ putative MH relevance
(We've released maps of all 77 networks in a new "Task-Evoked Network Atlas"): identifiers.org/neurovault.c...
(We've released maps of all 77 networks in a new "Task-Evoked Network Atlas"): identifiers.org/neurovault.c...
And it's not to pick on you guys or this paper - I know your group does fantastic, super high-quality work. It's just because this confound becomes especially pernicious in arousal change manipulations. I'll leave things there, but again happy to help if want to check this out in your PIDT sample
May 27, 2025 at 2:43 PM
And it's not to pick on you guys or this paper - I know your group does fantastic, super high-quality work. It's just because this confound becomes especially pernicious in arousal change manipulations. I'll leave things there, but again happy to help if want to check this out in your PIDT sample
Yes, pro physio signals (sLFO a physio signal too, detectable w/ finger pulse oximeter). But our core disagreement isn't about classic physio signals vs sLFO signal, it's about need to model these signals with their regional heterogeneity in mind (see tinyurl.com/442m7hnj); that's key of Riptide
May 27, 2025 at 2:43 PM
Yes, pro physio signals (sLFO a physio signal too, detectable w/ finger pulse oximeter). But our core disagreement isn't about classic physio signals vs sLFO signal, it's about need to model these signals with their regional heterogeneity in mind (see tinyurl.com/442m7hnj); that's key of Riptide
Where (in sensorimotor cortices) we’ve shown that changes in arousal have the largest effect on sLFO-driven FC distortion
May 24, 2025 at 8:16 PM
Where (in sensorimotor cortices) we’ve shown that changes in arousal have the largest effect on sLFO-driven FC distortion
That’s my point tho - it’s precisely stimulants’ affect on arousal (and therefore on the sLFO) that leads to my concern about the FC estimates being distorted, esp in the high vessel density areas of sensorimotor cortices
May 24, 2025 at 8:14 PM
That’s my point tho - it’s precisely stimulants’ affect on arousal (and therefore on the sLFO) that leads to my concern about the FC estimates being distorted, esp in the high vessel density areas of sensorimotor cortices
I wasn't involved in developing it and stand to gain nothing by anyone using it. I just think it's highly relevant to this dataset, and that there's strong reason to think it may clarify something important that standard physio regression may miss.
May 24, 2025 at 7:59 PM
I wasn't involved in developing it and stand to gain nothing by anyone using it. I just think it's highly relevant to this dataset, and that there's strong reason to think it may clarify something important that standard physio regression may miss.
Definitely agree there about stimulants increasing arousal
May 24, 2025 at 7:54 PM
Definitely agree there about stimulants increasing arousal
Oh for sure, definitely wasn't suggesting this for the ABCD dataset - but perhaps for some subset of the smaller replication dataset
May 24, 2025 at 7:53 PM
Oh for sure, definitely wasn't suggesting this for the ABCD dataset - but perhaps for some subset of the smaller replication dataset
This is cool and looks like it captures something similar!
May 24, 2025 at 7:50 PM
This is cool and looks like it captures something similar!
I just think if the basis for the claims of which functional domains are affected by stimulants is resting on which brain networks are changing, we'd want to be sure its actually the brain networks that are changing and not a noise signal that's just making it look like they're changing
May 24, 2025 at 6:44 PM
I just think if the basis for the claims of which functional domains are affected by stimulants is resting on which brain networks are changing, we'd want to be sure its actually the brain networks that are changing and not a noise signal that's just making it look like they're changing
The issue is that the sLFO signal is not only temporally variable, it's spatially variable within the brain. One of RIPTiDe's key features is that is performs voxel-specific regressions to deal with this, & its only able to do this bc it directly models the signal (which physio recordings don't)
May 24, 2025 at 6:37 PM
The issue is that the sLFO signal is not only temporally variable, it's spatially variable within the brain. One of RIPTiDe's key features is that is performs voxel-specific regressions to deal with this, & its only able to do this bc it directly models the signal (which physio recordings don't)