Machine Learning-Based Mobile Application for Predicting Posterior Canal Benign Paroxysmal Positional Vertigo.

Area of Science:

• Vestibular science
• Machine learning in healthcare
• Medical diagnostics

Background:

• Benign Paroxysmal Positional Vertigo (BPPV) is a common cause of vertigo.
• Accurate and timely diagnosis of PC-BPPV is crucial for effective treatment.
• Current diagnostic methods can be time-consuming and require specialized clinics.

Purpose of the Study:

• To investigate the predictability of Posterior Canal BPPV (PC-BPPV) using machine learning models based on vertigo/dizziness features and patient medical history.
• To develop a mobile application incorporating the highest-accuracy predictive model for PC-BPPV.

Main Methods:

• Retrospective analysis of medical records from 280 patients presenting with dizziness or vertigo.
• Utilized demographic information, medical history, and dizziness/vertigo characteristics.
• Eight machine learning models were evaluated, with the top-performing model integrated into a mobile application.

Main Results:

• Key distinguishing factors for PC-BPPV included age, symptom onset, duration, dizziness type, triggers, and auditory status.
• The Random Forest algorithm achieved the highest accuracy at 96.43% in predicting PC-BPPV.
• Other models demonstrated accuracies ranging from 89.28% to 94.64%.

Conclusions:

• Machine learning, leveraging dizziness characteristics and medical history, offers a highly accurate method for predicting PC-BPPV.
• The developed mobile application highlights the potential of AI in advancing telemedicine for vestibular disorders.