Pattern analysis of peripheral-vestibular deficits with machine learning using hierarchical clustering

Abstract

BackgroundDisorders affecting the vestibular organs (semicircular canals, utriculus, sacculus), may result in distinct patterns of peripheral-vestibular loss that may facilitate the diagnostic assessment. When neuropathological tests of these sensors are available, it is possible to classify responses as being due to different deficit types.ObjectiveTo provide a topical review and to summarize recent advances in pattern-recognition of unilateral and bilateral vestibular disease by use of hierarchical cluster analysis (HCA) as published by the authors.HypothesisWe propose that certain patterns of peripheral-vestibular loss are associated with specific underlying disorders and that HCA is a suitable approach to identify such patterns.DiscussionIn the studies reviewed, disease-specific patterns could be recognized in different patient cohorts, with anterior-canal sparing being a hallmark feature in aminoglycoside-related bilateral vestibulopathy, Menière's disease and vestibular Schwannoma. The reasons for such anterior-canal sparing remain subject to debate, but potential explanations include reduced toxic exposure, faster recovery and lower vulnerability of the anterior canals. The pattern observed in acute superior-branch vestibular neuropathy, i.e., involvement of the horizontal and anterior canal and the utricle, matches neural inner-ear physiology. The broadly varying extent of damage to the different vestibular sensors even within given disorders underlines the necessity for detailed vestibular-testing.ConclusionHCA significantly facilitates pattern-identification in unilateral and bilateral vestibulopathies and underlines the extensive range of vestibular end-organ damage in the different study populations and subgroups. The large number of existing clustering algorithms with distinct strengths and weaknesses emphasizes the need for careful selection of the most suitable algorithm.