Neuroanesthesia & neurocritical care
Application of a time series foundation model to noninvasively estimate intracranial pressure
Authors' conclusion
We present a framework for leveraging a time-series foundation model for deriving robust nICP measures across diverse patient populations.
Study context
- Population
- Brain-injured patients with intracranial pressure (ICP) monitoring from two cohorts: Columbia University Irving Medical Center (CUIMC) (n=11, median age 56 years) and MIMIC III (n=11, median age 11 years)11 CUIMC, 11 MIMIC
- Participants
- 11 vs 11
- Comparison
- CUIMC-trained model vs MIMIC-trained model
- CUIMC-trained modeln = 11
- MIMIC-trained modeln = 11
- Outcome
- Accuracy of non-invasive ICP estimation (median absolute error, Bland-Altman statistics)
Methodological detailsleave-one-patient-out validation, cross-cohort external validation, two cohorts
- Leave-one-patient-out validation
- Cross-cohort external validation
- Two cohorts
Performance
As reported in the abstract
Median absolute error of nICP
| Test or model | Value | Tested on |
|---|---|---|
| CUIMC-trained model | median 3.4 (IQR 2.54–5.69) mmHg | CUIMC dataset (within-cohort) |
| CUIMC-trained model | median 3.21 (IQR 2.46–3.81) mmHg | MIMIC dataset (cross-cohort) |
| MIMIC-trained model | median 3.99 (IQR 2.56–5.96) mmHg | MIMIC dataset (within-cohort) |
| MIMIC-trained model | median 1.97 (IQR 1.87–3.82) mmHg | CUIMC dataset (cross-cohort) |
Value
CUIMC-trained modelmedian 3.4 (IQR 2.54–5.69) mmHg
CUIMC-trained modelmedian 3.21 (IQR 2.46–3.81) mmHg
MIMIC-trained modelmedian 3.99 (IQR 2.56–5.96) mmHg
MIMIC-trained modelmedian 1.97 (IQR 1.87–3.82) mmHg
Tested on
CUIMC-trained modelCUIMC dataset (within-cohort)
CUIMC-trained modelMIMIC dataset (cross-cohort)
MIMIC-trained modelMIMIC dataset (within-cohort)
MIMIC-trained modelCUIMC dataset (cross-cohort)
Original publication
Application of a time series foundation model to noninvasively estimate intracranial pressure. J Clin Monit Comput. Published 9 October 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →