Auto-Pattern Recognition for Diagnosis in Benign Paroxysmal Positional Vertigo Using Principal Component Analysis: A Preliminary Study

Abstract

Objectives: The aim of this study was to develop a filtering algorithm for raw nystagmus images and a diagnostic assistive algorithm using a principal component analysis (PCA) to distinguish the different types of benign paroxysmal positional vertigo (BPPV).Methods: Fifteen video clips of clinical data with typical nystagmus patterns of BPPV (13 cases) and with normal nystamgmus (two cases) were preprocessed when applied the thresholding, morphology operation, residual noise filtering, and center point extraction stages. We analyzed multiple data clusters in a single frame via a PCA; in addition, we statistically analyzed the horizontal and vertical components of the main vector among the multiple data clusters in the canalolithiasis of the lateral semicircular canal (LSCC) and the posterior semicircular canal (PSCC).Results: We obtained a clear imaginary pupil and data on the fast phases and slow phases after preprocessing the images. For a normal patient, a round shape of clustered dots was observed. Patients with LSCC showed an elongated horizontal shape, whereas patients with PSCC showed an oval shape at the (x, y) coordinates. The scalar values (mm) of the horizontal component of the main vector when performing a PCA between the LSCC- and PSCC-BPPV were substantially different (102.08±20.11 vs. 32.36±12.52 mm, respectively; <i>p</i>=0.0012). Additionally, the salar ratio of horizontal to vertical components in LSCC and PSCC exhibited a significant difference (16.11±10.74 mm vs. 2.61±1.07 mm, respectively; <i>p</i>=0.0023).Conclusions: The data of a white simulated imaginary pupil without any background noise can be a separate monitoring option, which can aid clinicians in determining the types of BPPV exhibited. Therefore, this analysis algorithm will provide assistive information for diagnosis of BPPV to clinicians.