Detecting Atrial Fibrillation by Artificial Intelligence-Enabled Neuroimaging Examination.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Occult atrial fibrillation (AF) is often asymptomatic and difficult to diagnose, leading to under-detection and increased stroke risk.
• Current diagnostic methods for AF have limitations in feasibility and accuracy.
• Machine learning (ML) shows promise for improving clinical decision-making, particularly in analyzing complex medical data like magnetic resonance imaging (MRI).

Purpose of the Study:

• To develop and validate a machine learning algorithm for classifying stroke patients based on the presence or absence of atrial fibrillation (AF) using MRI data.
• To investigate the hypothesis that an ML algorithm can improve the accuracy of differentiating stroke etiology due to AF versus large artery atherosclerosis.

Main Methods:

• A cohort of 235 stroke patients (97 with AF, 138 without AF) was analyzed.
• Patients were randomly assigned to training (4:1 ratio) and validation groups.
• A 3D convolutional neural network (ConvNeXt) was developed and trained on MRI data for binary classification of stroke etiology.

Main Results:

• The ML model achieved an Area Under the Receiver Operating Characteristic Curve (AUC-ROC) of 0.88 in the best cross-validation fold, with an overall performance of 0.81 ± 0.05.
• The algorithm demonstrated strong performance with a precision of 0.84 ± 0.08 and an F1-score of 0.77 ± 0.06.
• The model showed reasonable classification power in distinguishing stroke patients with and without underlying AF.

Conclusions:

• The developed machine learning algorithm shows potential for accurately classifying stroke patients with and without underlying atrial fibrillation (AF) based on MRI.
• Further validation in external datasets is crucial to confirm the generalizability and clinical utility of this ML approach for AF detection in stroke patients.