aceCT-based prediction of hematoma expansion and adverse outcomes after spontaneous intracerebral hemorrhage
Source article: CT-based prediction of hematoma expansion and adverse outcomes after intracerebral hemorrhage: evidence appraisal, artificial intelligence translation, and GeroScience perspectives
Abstract: Spontaneous intracerebral hemorrhage (ICH) is a highly lethal and disabling form of stroke, in which hematoma expansion (HE) is a major and potentially modifiable determinant of early neurological deterioration and poor functional outcome. Computed tomography (CT) remains the first-line imaging modality for acute ICH and provides essential information for early HE risk stratification. However, current evidence is dispersed across conventional CT signs, composite scores, radiomics, machine learning, and deep lear…
Contested: both sides are scored from claims and sources, not community votes.
"Brain Microbleed" by National Institutes of Health (NIH) is marked with Public Domain Mark 1.0. To view the terms, visit https://creativecommons.org/publicdomain/mark/1.0/.
This peer-reviewed review in GeroScience appraises CT-based prediction of hematoma expansion after spontaneous intracerebral hemorrhage, a major determinant of early deterioration. It compares contrast-enhanced signs like spot, leakage and iodine signs with non-contrast signs including blend, black hole, island, satellite, hypodensity and swirl signs, plus shape and heterogeneity, and evaluates composite scores and AI approaches including radiomics, machine learning and deep learning.
The appraisal matters because hematoma expansion is potentially modifiable and CT is first-line imaging, so reliable prediction could guide individualized management. The authors find the literature fragmented and many AI models lacking external validation and workflow readiness, with additional complexity from vascular aging, amyloid angiopathy, frailty and anticoagulant exposure, leaving uncertainty about which markers or models will translate to bedside use.
- Review compares contrast-enhanced markers (spot sign, leakage sign, iodine sign) with non-contrast markers (blend sign, black hole sign, island sign, satellite sign, hypodensity sign, swirl sign, shape and density heterogeneity) on sensitivity, specificity, reproducibility and availability.
- Authors note current evidence is dispersed across conventional signs, composite scores, radiomics, machine learning and deep learning without sufficient comparison of clinical utility or validation quality.
- Discussion includes aging-related factors such as vascular aging, cerebral amyloid angiopathy, frailty, anticoagulant exposure and vulnerability to secondary injury influencing HE risk.
CT-based prediction, including AI approaches, provides essential information for early hematoma expansion risk stratification to support individualized management after spontaneous intracerebral hemorrhage.
Artificial intelligence approaches for CT-based prediction of hematoma expansion and adverse outcomes after spontaneous ICH are limited by small cohorts, overfitting, dataset heterogeneity, insufficient external validation, poor interpretability and lack of workflow integration.
The rundown
The review systematically contrasts contrast-enhanced CT markers with a set of non-contrast signs, evaluating each on sensitivity, specificity, reproducibility, availability and clinical applicability for early risk stratification.
It then appraises composite scores and AI methods, highlighting recurring methodological issues and calling for interpretable, multimodal, prospectively validated models that integrate imaging, clinical variables, biomarkers and aging-related factors.
Evidence base is fragmented and many AI models lack robust validation, with small cohorts, overfitting, dataset heterogeneity and insufficient external validation limiting translational readiness and workflow integration.
Sources
- Peer-reviewedGeroScience2026-08-15
How should this claim be treated?
The debate