Prognostische Bedeutung der KI-basierte Analyse intrazerebraler Blutungen unter Verwendung von Radiomics und klinischen Merkmalen bei intrazerebralen Blutungen.

Intracerebral hemorrhage (ICH) is associated with a significant disability-adjusted life years lost. Treatment decision-making largely dependents on clinical judgement of health care professionals lacking rigor and reproducibility. We hypothesized that that AI-based radiomics could guide unbiased prediction of hematoma expansion (HE) and clinical outcomes.

Patients with spontaneous ICB between 2015 – 2020, were enrolled in this large monocentric retrospective analysis. Modified Rankin Score (mRS) and Glasgow Outcome Scale Extended (GOSE) were used to determine functional outcome. 107 Radiomics features were extracted from non-contrast enhanced Computer Tomography (CT) slice images at the time of admission using 3D-Slicer (Open-Source-Python) together with PyRadiomics. Logistic regression, support vector machine, and random forest classifiers were built using the non-correlated selected radiomic, clinical, and dose features on the training dataset and performance was assessed in the testing dataset. The area under the curve (AUC) was used to assess the prognostic value.

202 patients with intracerebral hemorrhage (ICH) were analyzed. Median age was 76 years (34 – 93 years), female:male ratio was 94:108. Mean hematoma volume on admission was 37.2±15ml. Early HE within 6h occurred in 52 patients (26%), HE within 48h in 10 patients (5%). 30-day mortality was 26%. At discharge, median GOSE was 3 (1-7) and median mRS was 4 (0-6). Overall, 61 patients (30.4%) reached a favorable outcome (GOSE 5-8) at time of discharge. AI-based analysis using clinical and radiomics features achieved good prediction performance for HE (AUC 0.746±0.159; 95%CI (0.481-0.878)). Further, the combined model achieved good discrimination of poor outcomes at the time of discharge (AUC 0.975±0.003; 95%CI (0.923-0.999)).

AI-based assisted analysis of ICH based on clinical and radiomic features is capable of predicting impending HE and poor short-term outcome. Unbiased prediction of the clinical course and outcome might help to increase therapeutic efficacy, rigorous decision making and ultimately improve treatment of patients with ICH.