Accuracy of automated intracerebral hemorrhage volume measurement on non-contrast computed tomography: a Swedish Stroke Register cohort study.

PURPOSE: Hematoma volume is the strongest predictor of patient outcome after intracerebral hemorrhage (ICH). The aim of this study was to validate novel fully automated software for quantification of ICH volume on non-contrast computed tomography (CT). METHODS: The population was defined from the Swedish Stroke Register (RS) and included all patients with an ICH diagnosis during 2016-2019 in Region Skåne. Hemorrhage volume on their initial head CT was measured using ABC/2 and manual segmentation (Sectra IDS7 volume measurement tool) and the automated volume quantification tool (qER-NCCT) by Qure.ai. The first 500 were examined by two independent readers. RESULTS: A total of 1649 ICH patients were included. The qER-NCCT had 97% sensitivity in identifying ICH. In total, there was excellent agreement between volumetric measurements of ICH volumes by qER-NCCT and manual segmentation by interclass correlation (ICC = 0.96), and good agreement (ICC = 0.86) between qER-NCCT and ABC/2 method. The qER-NCCT showed volume underestimation, mainly in large (> 30 ml) heterogenous hemorrhages. Interrater agreement by (ICC) was 0.996 (95% CI: 0.99-1.00) for manual segmentation. CONCLUSION: Our study showed excellent agreement in volume quantification between the fully automated software qER-NCCT and manual segmentation of ICH on NCCT. The qER-NCCT would be an important additive tool by aiding in early diagnostics and prognostication for patients with ICH and in provide volumetry on a population-wide level. Further refinement of the software should address the underestimation of ICH volume seen in a portion of large, heterogenous, irregularly shaped ICHs.

Evaluation of CINA® LVO artificial intelligence software for detection of large vessel occlusion in brain CT angiography.

Objective: To systematically evaluate the ability of the CINA® LVO software to detect large vessel occlusions eligible for mechanical thrombectomy on CTA using conventional neuroradiological assessment as gold standard. Methods: Retrospectively, two hundred consecutive patients referred for a brain CTA and two hundred patients that had been subject for endovascular thrombectomy, with an accessible preceding CTA, were assessed for large vessel occlusions (LVO) using the CINA® LVO software. The patients were sub-grouped by occlusion site. The original radiology report was used as ground truth and cases with disagreement were reassessed. Two-by-two tables were created and measures for LVO detection were calculated. Results: A total of four-hundred patients were included; 221 LVOs were present in 215 patients (54 %). The overall specificity was high for LVOs in the anterior circulation (93 %). The overall sensitivity for LVOs in the anterior circulation was 54 % with the highest sensitivity for the M1 segment of the middle cerebral artery (87 %) and T-type internal carotid occlusions (84 %). The sensitivity was low for occlusions in the M2 segment of the middle cerebral artery (13 % and 0 % for proximal and distal M2 occlusions respectively) and in posterior circulation occlusions (0 %, not included in the intended use of the software). Conclusions: LVO detection sensitivity for the CINA® LVO software differs largely depending on the location of the occlusion, with low sensitivity for detection of some LVOs potentially eligible for mechanical thrombectomy. Further development of the software to increase sensitivity to all LVO locations would increase the clinical usefulness.

Radiological features of brain hemorrhage through automated segmentation from computed tomography in stroke and traumatic brain injury.

Introduction: Radiological assessment is necessary to diagnose spontaneous intracerebral hemorrhage (ICH) and traumatic brain injury intracranial hemorrhage (TBI-bleed). Artificial intelligence (AI) deep learning tools provide a means for decision support. This study evaluates the hemorrhage segmentations produced from three-dimensional deep learning AI model that was developed using non-contrast computed tomography (CT) imaging data external to the current study. Methods: Non-contrast CT imaging data from 1263 patients were accessed across seven data sources (referred to as sites) in Norway and Sweden. Patients were included based on ICH, TBI-bleed, or mild TBI diagnosis. Initial non-contrast CT images were available for all participants. Hemorrhage location frequency maps were generated. The number of estimated haematoma clusters was correlated with the total haematoma volume. Ground truth expert annotations were available for one ICH site; hence, a comparison was made with the estimated haematoma volumes. Segmentation volume estimates were used in a receiver operator characteristics (ROC) analysis for all samples (i.e., bleed detected) and then specifically for one site with few TBI-bleed cases. Results: The hemorrhage frequency maps showed spatial patterns of estimated lesions consistent with ICH or TBI-bleed presentations. There was a positive correlation between the estimated number of clusters and total haematoma volume for each site (correlation range: 0.45-0.74; each p-value < 0.01) and evidence of ICH between-site differences. Relative to hand-drawn annotations for one ICH site, the VIOLA-AI segmentation mask achieved a median Dice Similarity Coefficient of 0.82 (interquartile range: 0.78 and 0.83), resulting in a small overestimate in the haematoma volume by a median of 0.47 mL (interquartile range: 0.04 and 1.75 mL). The bleed detection ROC analysis for the whole sample gave a high area-under-the-curve (AUC) of 0.92 (with sensitivity and specificity of 83.28% and 95.41%); however, when considering only the mild head injury site, the TBI-bleed detection gave an AUC of 0.70. Discussion: An open-source segmentation tool was used to visualize hemorrhage locations across multiple data sources and revealed quantitative hemorrhage site differences. The automated total hemorrhage volume estimate correlated with a per-participant hemorrhage cluster count. ROC results were moderate-to-high. The VIOLA-AI tool had promising results and might be useful for various types of intracranial hemorrhage.

Computed tomography in acute intracerebral hemorrhage: neuroimaging predictors of hematoma expansion and outcome.

Intracerebral hemorrhage (ICH) accounts for 10-20% of all strokes worldwide and is associated with serious outcomes, including a 30-day mortality rate of up to 40%. Neuroimaging is pivotal in diagnosing ICH as early detection and determination of underlying cause, and risk for expansion/rebleeding is essential in providing the correct treatment. Non-contrast computed tomography (NCCT) is the most used modality for detection of ICH, identification of prognostic markers and measurements of hematoma volume, all of which are of major importance to predict outcome. The strongest predictors of 30-day mortality and functional outcome for ICH patients are baseline hematoma volume and hematoma expansion. Even so, exact hematoma measurement is rare in clinical routine practice, primarily due to a lack of tools available for fast, effective, and reliable volumetric tools. In this educational review, we discuss neuroimaging findings for ICH from NCCT images, and their prognostic value, as well as the use of semi-automatic and fully automated hematoma volumetric methods and assessment of hematoma expansion in prognostic studies.