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

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science in Medicine

Background:

• Stroke, encompassing ischemic stroke (AIS) and intracerebral hemorrhage (ICH), presents a significant global health challenge.
• Distinct management strategies are required for AIS and ICH, making early and accurate differentiation critical for patient outcomes.
• Current diagnostic methods predominantly rely on imaging techniques like CT and MRI, which may not be readily available in all settings.

Purpose of the Study:

• To investigate a non-imaging approach for differentiating between AIS and ICH.
• To evaluate the efficacy of a combined Principal Component Analysis (PCA) and Logistic Regression (LR) model using physiological parameters.
• To assess the potential of this method for ultra-acute stroke care and prehospital settings.

Main Methods:

• A retrospective analysis of 80 mild-to-moderate stroke patients (68 AIS, 12 ICH) was conducted.
• Sixty-seven parameters, including baroreceptor sensitivity (BRS) and hemodynamic variables, were analyzed using PCA and LR.
• Model performance was validated using two-fold and six-fold cross-validation techniques.

Main Results:

• The PCA-LR model successfully differentiated between AIS and ICH.
• BRS parameters and cerebral hemodynamic factors were identified as significant contributors to diagnostic accuracy.
• Two-fold cross-validation yielded an Area Under the Curve (AUC) of ≥0.92, and six-fold cross-validation achieved AUC ≥0.79.

Conclusions:

• A non-imaging, multidimensional approach using physiological data can effectively differentiate between AIS and ICH.
• This method offers a potential tool for rapid stroke subtype identification, particularly in prehospital environments.
• Further research with larger datasets is necessary to validate clinical applicability and address limitations, such as distinguishing stroke from mimics.