[The neural network algorithm for diagnosis of ischemic stroke pathogenetic subtypes]

Area of Science:

• Neurology
• Medical Informatics
• Biomedical Engineering

Background:

• Accurate ischemic stroke (IS) subtyping is crucial for effective treatment.
• Traditional diagnostic methods and algorithms like TOAST have limitations in classifying IS subtypes.
• Expert clinical assessment is the standard but can be time-consuming and subjective.

Purpose of the Study:

• To develop and validate an artificial neural network (ANN) algorithm for determining IS pathogenetic subtypes.
• To compare the ANN algorithm's diagnostic performance against expert classification and the TOAST algorithm.
• To improve the accuracy and efficiency of IS subtype diagnosis.

Main Methods:

• Retrospective analysis of 204 acute ischemic stroke cases.
• Clinical evaluation, neuroimaging (CT), vascular imaging (duplex scanning), cardiac assessment (ECG, echocardiography), and Holter monitoring were utilized.
• Development of an ANN algorithm to estimate probabilities for atherothrombotic, cardioembolic, and lacunar IS subtypes.

Main Results:

• The TOAST algorithm demonstrated a high failure rate (73%) in subtype diagnosis.
• The developed ANN algorithm achieved high diagnostic accuracy.
• Key performance metrics for the ANN included 97% sensitivity, 98% positive predictive value, and an excellent index of agreement with expert diagnosis (K=0.955).

Conclusions:

• Artificial neural networks offer a promising, highly accurate tool for ischemic stroke subtype classification.
• The ANN algorithm significantly outperforms the TOAST algorithm and approaches expert-level diagnostic agreement.
• This AI-driven approach has the potential to enhance clinical decision-making and personalize stroke management.