Functional neuroimaging features for predicting the transition from benign paroxysmal positional vertigo to persistent postural-perceptual dizziness.
Authors
Affiliations (3)
Affiliations (3)
- Department of Geriatrics, Xijing Hospital, Air Force Medical University, Xi'an, China.
- Department of Neurology, Xijing Hospital, Air Force Medical University, Xi'an, China.
- Department of Anatomy, Histology and Embryology and K. K. Leung Brain Research Centre, Air Fourth Medical University, Xi'an, China.
Abstract
ObjectivesBenign paroxysmal positional vertigo (BPPV) is a prevalent triggers of persistent postural-perceptual dizziness (PPPD). The maladaptation of brain function may be one of the pathophysiology in PPPD. This study aims to identify brain functional neuroimaging features and establish prediction models to predict PPPD after BPPV.MethodsThe diagnosis of BPPV and PPPD was based on the criteria established by the Bárány Society. Patients with posterior semicircular canal BPPV were treated using the Epley maneuver. Patients with geotropic lateral canal BPPV were treated with the barbecue rotation maneuver, while those with apogeotropic lateral canal BPPV were treated using the Gufoni maneuver. After successful canalith repositioning maneuver treatment, the patient underwent resting-state functional magnetic resonance imaging (fMRI) scan. Using feature selection and extraction techniques, six machine learning algorithms were implemented to predict PPPD. The models were trained with 5-fold cross-validation, and performance was evaluated using the receiver operating characteristic curve (AUC), accuracy, precision, recall, and F1 score (F1).ResultsA total of 101 patients were included in the final analysis, comprising 64 patients without PPPD (non-PPPD) and 37 patients with PPPD (PPPD). A total of 22 functional neuroimaging features were identified to be closely associated with PPPD after BPPV. Among the six machine learning algorithms, the Multilayer Perceptron model exhibited superior performance, with an AUC of 0.93, a recall of 0.82, a precision of 0.83, an accuracy of 0.82, and an F1 score of 0.82. SHAP analysis identified the most influential resting-state fMRI features in this model. For the top 10 important resting-state fMRI features, 3 features overlapped in all six machine learning algorithms. These features include FC between the vermis 3 and the superior frontal gyrus, orbital part, DC in the cerebellum 7b, left, and FC between the Heschl gyrus, left, and the caudate, right.ConclusionsThese findings provide brain functional neuroimaging features which may be closely associated with the transition from BPPV to PPPD, thereby offering a valuable tool for the early detection of PPPD.