Can we identify stroke sub-type without imaging? A multidimensional analysis

Stroke is a major cause of mortality and disability worldwide, with ischemic stroke (AIS) and intracerebral haemorrhage (ICH) requiring distinct management approaches. Accurate early detection and differentiation of these subtypes is crucial for targeted treatment and improved patient outcomes. Traditionally, imaging techniques such as computed tomography (CT) and magnetic resonance imaging (MRI) are required to distinguish between AIS and ICH. However, this study explores a non-imaging approach to differentiate between stroke subtypes. Using a retrospective dataset of 80 mild-to-moderate patients suffering stroke (68 AIS and 12 ICH), we employed principal component analysis (PCA) combined with logistic regression (LR) to evaluate 67 parameters. These parameters include baroreceptor sensitivity, and cerebral and peripheral hemodynamic variables. The PCA-LR model, validated through two-fold and six-fold cross-validation methods, effectively differentiated between AIS and ICH. BRS parameters and cerebral hemodynamic factors contributed significantly to the model's accuracy. The two-fold cross-validation approach achieved an area under the curve (AUC) of ≥0.92, while the six-fold method maintained a consistent variance explanation (AUC ≥0.79). Results suggest that this multidimensional approach may facilitate early stroke subtype identification (AIS vs ICH) without reliance on imaging, offering a promising tool for ultra-acute stroke care in prehospital settings. However, it is important to note that the model has been tested in confirmed stroke cases, and its ability to distinguish between stroke and stroke mimics remains an important limitation for broader clinical application. Future research with larger datasets is warranted to refine the model and validate its clinical applicability.