A retrospective analysis of brain computed tomography scans has identified lesion dimensions as the primary imaging feature independently linked to three-month recovery trajectories following spontaneous intracerebral haemorrhage. Researchers Murat Özdemir and Fatma Latifoğlu examined the prognostic utility of quantitative density measurements and structural dimensions extracted from admission scans. While lower-tail attenuation values initially appeared relevant in basic statistical comparisons, only structural lesion dimensions demonstrated both multivariable predictive value and reliable measurement consistency across independent assessments.
Spontaneous intracerebral haemorrhage constitutes a severe medical condition characterized by substantial patient mortality and frequent long-term functional impairment. Although computed tomography remains a foundational diagnostic tool upon hospital admission, the clinical relevance and reliability of density-based Hounsfield Unit metrics have remained controversial across neuroimaging literature. The exploratory investigation specifically sought to clarify whether region-specific measurements focusing on lower Hounsfield Unit distributions and low-attenuation burden could reliably forecast patient outcomes alongside traditional clinical indicators.
The study cohort comprised 132 patients admitted with spontaneous intracerebral haemorrhage whose functional status was formally evaluated three months later using the modified Rankin Scale. The investigators stratified the patient population into two distinct functional categories: a favorable outcome group comprising 69 individuals with scores from 0 to 3, and an unfavorable outcome group encompassing 63 individuals with scores ranging from 4 to 6. Baseline imaging data gathered during initial hospital presentations served as the core dataset for subsequent computational evaluation and feature extraction.
Image processing involved semi-automatic segmentation of admission computed tomography slices into three distinct anatomical compartments: the haematoma core, the surrounding perihematomal oedema, and a combined core-plus-oedema region. From each defined compartment, the research team extracted morphological parameters alongside detailed density-based metrics. The lower-tail density analysis focused specifically on metrics designated as p01, ratio_low10, and mean_low20, designed to capture varying levels of low-attenuation burden within the damaged brain tissue.
Statistical evaluation of group differences relied on the Mann–Whitney U test and Cohen’s d, complemented by Benjamini–Hochberg false discovery rate corrections applied across ten pre-specified analytical features within each segmented region. Initial univariate assessments indicated that larger lesion extents and greater low-attenuation burdens correlated with poor patient outcomes. Morphological indicators, specifically lesion area and minor axis length, exhibited statistically significant differences between outcome groups across all segmented anatomical compartments.
Following false discovery rate adjustment within each region, five specific parameters within the perihematomal oedema compartment maintained statistically significant univariate associations with functional recovery: minor axis length, minimum Hounsfield Units, p01, ratio_low10, and mean_low20, with adjusted p-values ranging from 0.045 to 0.046. Conversely, standard median Hounsfield Unit measurements failed to display any consistent relationship with functional endpoints, underscoring the limited utility of central density markers in this diagnostic setting.
A critical component of the methodology involved testing intra-observer measurement reproducibility across two independent repeat-segmentation cohorts utilizing intraclass correlation coefficients. While structural dimensions demonstrated excellent reproducibility, the lower-tail density metrics ratio_low10 and mean_low20 yielded poor or inconsistent reproducibility scores. This instability across repeated evaluations highlighted a fundamental limitation in the reliability of nuanced density measurements when applied across semi-automated image segmentations.
Multivariable logistic regression modeling was subsequently conducted to adjust for standard clinical risk factors, including patient age, Glasgow Coma Scale score on admission, and haematoma location. Within this adjusted framework, minor axis length within the perihematomal oedema region remained an independent predictor of poor outcome, yielding an odds ratio of 1.02 with a 95 percent confidence interval of 1.00 to 1.04. Incorporating this structural marker improved overall model discrimination beyond clinical parameters alone, increasing the area under the receiver operating characteristic curve from 0.82 to 0.84.
The authors concluded that morphological lesion extent represented the sole feature supported by both independent prognostic association and robust measurement reproducibility. Although lower-tail density features showed preliminary univariate correlations, they offered no verified incremental prognostic utility beyond established markers. Given that the investigation was an exploratory, single-centre study reliant on single-slice computed tomography assessments, the authors emphasized that prospective volumetric analyses in larger, externally validated cohorts remain necessary before these imaging parameters can guide clinical decision-making.
Subject of Research: Prognostic evaluation of CT lesion size and attenuation features in spontaneous intracerebral haemorrhage
Article Title: Region-specific lesion size and low-HU burden as prognostic markers in intracerebral haemorrhage: an exploratory CT analysis
Article References: Özdemir, M., & Latifoğlu, F. (2026). Region-specific lesion size and low-HU burden as prognostic markers in intracerebral haemorrhage: an exploratory CT analysis. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-03015-6
Image Credits: AI Generated
DOI: 10.1186/s12880-026-03015-6
Keywords: intracerebral haemorrhage, computed tomography, perihematomal oedema, haematoma core, Hounsfield Unit, modified Rankin Scale, prognostic markers, neuroimaging, lesion size, low-HU burden, functional outcome, image segmentation, intra-observer reproducibility, multivariable logistic regression, area under the curve, Glasgow Coma Scale, retrospective study
News Source: Ophelia Keating. (November 13, 2026). CT Lesion Size Outperforms Density Metrics in Predicting Haemorrhage Outcomes. Scienmag.



