Classification of vasovagal syncope from physiological signals on tilt table testing

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Vasovagal syncope (VVS) is a common cause of fainting, often diagnosed using the head-up tilt test (HUTT).
• HUTT involves monitoring physiological responses like blood pressure (BP) and electrocardiography (ECG) during postural changes.
• Patients may experience symptoms such as nausea, sweating, pallor, palpitations, and syncope during the test.

Purpose of the Study:

• To develop and evaluate a machine learning algorithm for classifying VVS patients.
• To utilize physiological signals (BP and ECG) obtained during HUTT for VVS diagnosis.
• To identify crucial features for accurate VVS classification.

Main Methods:

• Subjects underwent a 70-degree head-up tilt test with glyceryl trinitrate administration.
• Missing data were handled using K-nearest neighbors (KNN) imputation.
• Feature selection techniques and machine learning models (SVM, GNB, MNB, KNN, LR, RF) were employed for classification.
• Partial dependence plots were used for model interpretation.

Main Results:

• The study included 137 subjects (54 VVS positive, 83 negative).
• An optimal model combining KNN imputation, three tilting features, and Support Vector Machine (SVM) achieved 90.5% accuracy.
• Performance metrics included 87.0% sensitivity, 92.7% specificity, 88.6% precision, 87.8% F1 score, and 95.4% ROC AUC.

Conclusions:

• The developed algorithm effectively classifies VVS patients with high accuracy (>90%).
• The findings suggest a potential for improved VVS diagnosis using physiological signals and machine learning.
• Further validation with larger clinical datasets is recommended for generalizability.