Introduction: The diversity of underlying causes poses a medical challenge in the differential diagnosis of dizziness. For many patients, the time to reach a diagnosis is quite long and includes several tests. This imposes a significant financial burden on the healthcare system while delaying the recognition and treatment of some neurological emergencies. In this paper, we present a machine learning-based method to aid in the early recognition of vascular causes of vertigo, which require timely management.
Methods: Using data such as age, gender, accompanying symptoms, comorbidities, and commonly known blood parameters, a machine learning-based preliminary evaluation method was designed to predict the most appropriate group for the patient. After identifying the effective features using statistical methods, various machine learning methods (decision tree, logistic regression, support vector machines, k-nearest neighbor, multilayer perceptron, and ensemble learning methods) were employed to determine dizziness groups.
Results: Experimental results present that age, serum albumin, presence of headache, hypertension and diabetes are crucial features to classify vascular vertigo patients.Accuracies range from 81.7% to 86% and the best result is achieved with the logistic regression method. The decision tree, support vector machines, k-nearest neighbor, multilayer perceptron, and ensemble methods reached accuracies of 83%, 85.5%, 84%, 81.7%, and 82.8%, respectively.
Conclusion: Our study provides reasonable evidence suggesting that machine learning models may be useful in predicting vascular vertigo cases before hospital admission. Further studies are needed to confirm these findings and improve accuracy.Our model may be beneficial for ambulance personnel, practitioners and for the specialists facing atypical and difficult cases of dizziness.
Keywords: Artificial intelligence, dizziness, machine learning, stroke,