Background: Intracerebral hemorrhages (ICH) and perihematomal edema (PHE) are respective imaging markers of primary and secondary brain injury in hemorrhagic stroke. In this study, we explored the potential added value of PHE radiomic features for prognostication in ICH patients. Methods: Using a multicentric trial cohort of acute supratentorial ICH (n = 852) patients, we extracted radiomic features from ICH and PHE lesions on admission non-contrast head CTs. We trained and tested combinations of different machine learning classifiers and feature selection methods for prediction of poor outcome—defined by 4-to-6 modified Rankin Scale scores at 3-month follow-up—using five different input strategies: (a) ICH radiomics, (b) ICH and PHE radiomics, (c) admission clinical predictors of poor outcomes, (d) ICH radiomics and clinical variables, and (e) ICH and PHE radiomics with clinical variables. Models were trained on 500 patients, tested, and compared in 352 using the receiver operating characteristics Area Under the Curve (AUC), Integrated Discrimination Index (IDI), and Net Reclassification Index (NRI). Results: Comparing the best performing models in the independent test cohort, both IDI and NRI demonstrated better individual-level risk assessment by addition of PHE radiomics as input to ICH radiomics (both p < 0.001), but with insignificant improvement in outcome prediction (AUC of 0.74 versus 0.71, p = 0.157). The addition of ICH and PHE radiomics to clinical variables also improved IDI and NRI risk-classification (both p < 0.001), but with a insignificant increase in AUC of 0.85 versus 0.83 (p = 0.118), respectively. All machine learning models had greater or equal accuracy in outcome prediction compared to the widely used ICH score. Conclusions: The addition of PHE radiomics to hemorrhage lesion radiomics, as well as radiomics to clinical risk factors, can improve individual-level risk assessment, albeit with an insignificant increase in prognostic accuracy. Machine learning models offer quantitative and immediate risk stratification—on par with or more accurate than the ICH score—which can potentially guide patients’ selection for interventions such as hematoma evacuation.

Intracerebral hemorrhage (ICH) and perihematomal edema (PHE) are key imaging markers of primary and secondary brain injury in hemorrhagic stroke. Accurate segmentation and quantification of ICH and PHE can help with prognostication and guide treatment planning. In this study, we combined Swin-Unet Transformers with nnU-NETv2 convolutional network for segmentation of ICH and PHE on non-contrast head CTs. We also applied test-time data augmentations to assess individual-level prediction uncertainty, ensuring high confidence in prediction. The model was trained on 1782 CT scans from a multicentric trial and tested in two independent datasets from Yale (n = 396) and University of Berlin Charité Hospital and University Medical Center Hamburg-Eppendorf (n = 943). Model performance was evaluated with the Dice coefficient and Volume Similarity (VS). Our dual Swin-nnUNET model achieved a median (95% confidence interval) Dice = 0.93 (0.90–0.95) and VS = 0.97 (0.95–0.98) for ICH, and Dice = 0.70 (0.64–0.75) and VS = 0.87 (0.80–0.93) for PHE segmentation in the Yale cohort. Dice = 0.86 (0.80–0.90) and VS = 0.91 (0.85–0.95) for ICH and Dice = 0.65 (0.56–0.70) and VS = 0.86 (0.77–0.93) for PHE segmentation in the Berlin/Hamburg-Eppendorf cohort. Prediction uncertainty was associated with lower segmentation accuracy, smaller ICH/PHE volumes, and infratentorial location. Our results highlight the benefits of a dual transformer-convolutional neural network architecture for ICH/PHE segmentation and test-time augmentation for uncertainty quantification.

Hematoma expansion (HE) is an independent predictor of poor outcomes and a modifiable treatment target in intracerebral hemorrhage (ICH). Evaluating HE in large datasets requires segmentation of hematomas on admission and follow-up CT scans, a process that is time-consuming and labor-intensive in large-scale studies. Automated segmentation of hematomas can expedite this process; however, cumulative errors from segmentation on admission and follow-up scans can hamper accurate HE classification. In this study, we combined a tandem deep-learning classification model with automated segmentation to generate probability measures for false HE classifications. With this strategy, we can limit expert review of automated hematoma segmentations to a subset of the dataset, tailored to the research team’s preferred sensitivity or specificity thresholds and their tolerance for false-positive versus false-negative results. We utilized three separate multicentric cohorts for cross-validation/training, internal testing, and external validation (n = 2261) to develop and test a pipeline for automated hematoma segmentation and to generate ground truth binary HE annotations (≥3, ≥6, ≥9, and ≥12.5 mL). Applying a 95% sensitivity threshold for HE classification showed a practical and efficient strategy for HE annotation in large ICH datasets. This threshold excluded 47–88% of test-negative predictions from expert review of automated segmentations for different HE definitions, with less than 2% false-negative misclassification in both internal and external validation cohorts. Our pipeline offers a time-efficient and optimizable method for generating ground truth HE classifications in large ICH datasets, reducing the burden of expert review of automated hematoma segmentations while minimizing misclassification rate.

The National Institutes of Health (NIH) Toolbox cognition battery and Trail Making Tests (TMT) are widely used to quantify cognitive aging and to detect early cognitive vulnerability in Alzheimer’s disease and related conditions. However, these tests are often treated as interchangeable markers of global cognition, despite likely differences in their dependence on specific brain systems, limiting interpretability across studies and clinical contexts. To address this gap, we examined associations between four commonly used cognitive measures—fluid cognition, crystallized cognition, TMT-A, and TMT-B—and multimodal MRI metrics in 725 healthy volunteers aged 36 to 100 years from the Human Connectome Project–Aging. Voxel-wise diffusion MRI and vertex-wise cortical thickness and volume analyses were adjusted for age, sex, and years of education. Higher crystallized and fluid cognition scores and faster TMT-A/B completion times were generally associated with greater white matter integrity. TMT-B showed the most extensive diffusion and cortical associations, involving major projection, commissural, and association pathways and frontoparietal and temporo-occipital cortices. TMT-A and crystallized cognition demonstrated intermediate, overlapping patterns, whereas fluid cognition showed only focal brainstem and limited cortical correlates. These findings demonstrate systematic differences in the neuroanatomical substrates underlying commonly used cognitive tests and provide normative structure–cognition reference maps that can improve test selection, mechanistic interpretation, and sensitivity to brain health in studies of aging, vascular risk, and preclinical neurodegenerative disease.

Background: Early neurological deterioration (END) after intracerebral hemorrhage (ICH) is associated with poor prognosis, but its independent contribution to mortality and disability beyond baseline severity remains undetermined. Using causal inference methods, we estimated the effect of END on 3-month outcomes and the population-level impact of preventing END. Methods: Consecutive patients with acute ICH presenting within 12 h of symptom onset at a comprehensive stroke center (2014–2023) were retrospectively analyzed. END was defined as a ≥4-point increase in National Institutes of Health Stroke Scale (NIHSS) score or a ≥2-point decrease in Glasgow Coma Scale (GCS) score within 24 h. Multivariable logistic regression identified baseline predictors of END. The average treatment effect (ATE) of END on the 3-month mortality rate and poor functional outcome rate (mRS of 4–6) was estimated using G-computation, inverse probability weighting, and targeted maximum likelihood estimation (TMLE; primary), adjusting for baseline severity and clinical confounders. TMLE-based counterfactual analyses estimated the population-level impact of preventing END. Results: Among 690 patients (median age of 71 [interquartile: 60–81] years; 308 (44.6%) female), 273 (39.6%) developed END. Across the estimators, END increased the absolute risk of 3-month mortality by 23.1–23.9% and poor functional outcomes by 20.2–22.8%, with no significant effect on disability among the survivors. Counterfactual analyses projected that preventing END would reduce the mortality rate from 31.6% to 21.3% (population attributable fraction of 32.5%) and poor functional outcomes from 56.8% to 48.8% (population attributable fraction of 14.2%). Conclusions: In this single-center cohort, END was associated with a substantially higher 3-month mortality rate and poor functional outcomes, independent of measured baseline severity and other clinical confounders. If this is causal, then END is a potential target for early intervention and a candidate early efficacy endpoint for ICH trials; whether preventing END improves outcomes requires prospective evaluation.