Vestibular contribution to path integration deficits in 'at-genetic-risk' for Alzheimer's disease

Abstract

Path integration changes may precede a clinical presentation of Alzheimer's disease by several years. Studies to date have focused on how spatial cell changes affect path integration in preclinical AD. However, vestibular input is also critical for intact path integration. Here, we developed the vestibular rotation task that requires individuals to manually point an iPad device in the direction of their starting point following rotational movement, without any visual cues. Vestibular features were derived from the sensor data using feature selection. Machine learning models illustrate that the vestibular features accurately classified Apolipoprotein E ε3ε4 carriers and ε3ε3 carrier controls (mean age 62.7 years), with 65% to 79% accuracy depending on task trial. All machine learning models produced a similar classification accuracy. Our results demonstrate the cross-sectional role of the vestibular system in Alzheimer's disease risk carriers. Future investigations should examine if vestibular functions explain individual phenotypic heterogeneity in path integration among Alzheimer's disease risk carriers.

Fig 1. Semantic representation of the vestibular rotation task.

Fig 2. Machine learning models and feature importance.

A F1 scores for the best performing algorithm are shown. A random predictor would score 0.57, with a score of above 0.57 representing better-than-chance APOE status classification performance. Blue line includes all features. Red line excludes the path integration feature, end error. B Importance scores are represented by the circle diameter and were derived for the best performing model on each of the trials shown. Scores vary between 0 and 1 depending on the proportion of influence the feature has for that trial. RF = Random Forest, SVM = Support Vector Machine, MLP = Multi-Layer Perception.