Machine learning to predict penumbra core mismatch in acute ischemic stroke using clinical note data

Abstract In acute ischemic stroke due to large-vessel occlusion (AIS-LVO), late-window endovascular thrombectomy (EVT) decisions depend on penumbra-to-core (P:C) mismatch from computed tomographic perfusion (CTP). We developed multiple machine learning (ML) models to predict P:C ratios from a retros...

Full description

Saved in:
Bibliographic Details
Main Authors: Shaun Kohli, Parul Agarwal, “Andy” Ho Wing Chan, Asala Erekat, Girish Nadkarni, Benjamin Kummer
Format: Article
Language:English
Published: Nature Portfolio 2025-06-01
Series:npj Digital Medicine
Online Access:https://doi.org/10.1038/s41746-025-01703-1
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850136388143415296
author Shaun Kohli
Parul Agarwal
“Andy” Ho Wing Chan
Asala Erekat
Girish Nadkarni
Benjamin Kummer
author_facet Shaun Kohli
Parul Agarwal
“Andy” Ho Wing Chan
Asala Erekat
Girish Nadkarni
Benjamin Kummer
author_sort Shaun Kohli
collection DOAJ
description Abstract In acute ischemic stroke due to large-vessel occlusion (AIS-LVO), late-window endovascular thrombectomy (EVT) decisions depend on penumbra-to-core (P:C) mismatch from computed tomographic perfusion (CTP). We developed multiple machine learning (ML) models to predict P:C ratios from a retrospectively-identified cohort of AIS-LVO patients who underwent CTP within 30 min of initial neuroimaging, using non-imaging electronic health record (EHR) data available prior to CTP evaluation. We extracted structured data and free-text clinical notes from the EHR, generating document embeddings as sums of BioWordVec vectors weighted by term-frequency-inverse-document-frequency scores. We identified 120 patients; an extreme-gradient-boosting model classified P:C ratios as ≥ or <1.8, achieving an AUROC of 0.80 (95% CI 0.57–0.92) with optimal performance using text limited to 500 characters. Sensitivity was 0.80, specificity 0.66, and F1 score 0.86. Our findings suggest that ML models leveraging real-world non-imaging data can potentially aid LVO-AIS triage, though further validation is needed.
format Article
id doaj-art-2bb6b784d2364a4bb2fc30113c112da0
institution OA Journals
issn 2398-6352
language English
publishDate 2025-06-01
publisher Nature Portfolio
record_format Article
series npj Digital Medicine
spelling doaj-art-2bb6b784d2364a4bb2fc30113c112da02025-08-20T02:31:09ZengNature Portfolionpj Digital Medicine2398-63522025-06-01811810.1038/s41746-025-01703-1Machine learning to predict penumbra core mismatch in acute ischemic stroke using clinical note dataShaun Kohli0Parul Agarwal1“Andy” Ho Wing Chan2Asala Erekat3Girish Nadkarni4Benjamin Kummer5Icahn School of Medicine at Mount SinaiDepartment of Neurology, Icahn School of Medicine at Mount SinaiDepartment of Neurology, Icahn School of Medicine at Mount SinaiDepartment of Neurology, Icahn School of Medicine at Mount SinaiWindreich Department of Artificial Intelligence and Human Health, Icahn School of Medicine at Mount SinaiDepartment of Neurology, Icahn School of Medicine at Mount SinaiAbstract In acute ischemic stroke due to large-vessel occlusion (AIS-LVO), late-window endovascular thrombectomy (EVT) decisions depend on penumbra-to-core (P:C) mismatch from computed tomographic perfusion (CTP). We developed multiple machine learning (ML) models to predict P:C ratios from a retrospectively-identified cohort of AIS-LVO patients who underwent CTP within 30 min of initial neuroimaging, using non-imaging electronic health record (EHR) data available prior to CTP evaluation. We extracted structured data and free-text clinical notes from the EHR, generating document embeddings as sums of BioWordVec vectors weighted by term-frequency-inverse-document-frequency scores. We identified 120 patients; an extreme-gradient-boosting model classified P:C ratios as ≥ or <1.8, achieving an AUROC of 0.80 (95% CI 0.57–0.92) with optimal performance using text limited to 500 characters. Sensitivity was 0.80, specificity 0.66, and F1 score 0.86. Our findings suggest that ML models leveraging real-world non-imaging data can potentially aid LVO-AIS triage, though further validation is needed.https://doi.org/10.1038/s41746-025-01703-1
spellingShingle Shaun Kohli
Parul Agarwal
“Andy” Ho Wing Chan
Asala Erekat
Girish Nadkarni
Benjamin Kummer
Machine learning to predict penumbra core mismatch in acute ischemic stroke using clinical note data
npj Digital Medicine
title Machine learning to predict penumbra core mismatch in acute ischemic stroke using clinical note data
title_full Machine learning to predict penumbra core mismatch in acute ischemic stroke using clinical note data
title_fullStr Machine learning to predict penumbra core mismatch in acute ischemic stroke using clinical note data
title_full_unstemmed Machine learning to predict penumbra core mismatch in acute ischemic stroke using clinical note data
title_short Machine learning to predict penumbra core mismatch in acute ischemic stroke using clinical note data
title_sort machine learning to predict penumbra core mismatch in acute ischemic stroke using clinical note data
url https://doi.org/10.1038/s41746-025-01703-1
work_keys_str_mv AT shaunkohli machinelearningtopredictpenumbracoremismatchinacuteischemicstrokeusingclinicalnotedata
AT parulagarwal machinelearningtopredictpenumbracoremismatchinacuteischemicstrokeusingclinicalnotedata
AT andyhowingchan machinelearningtopredictpenumbracoremismatchinacuteischemicstrokeusingclinicalnotedata
AT asalaerekat machinelearningtopredictpenumbracoremismatchinacuteischemicstrokeusingclinicalnotedata
AT girishnadkarni machinelearningtopredictpenumbracoremismatchinacuteischemicstrokeusingclinicalnotedata
AT benjaminkummer machinelearningtopredictpenumbracoremismatchinacuteischemicstrokeusingclinicalnotedata