Inertial Measurement of Head Tilt in Rodents: Principles and Applications to Vestibular Research

Machine Learning Analysis of Physical Activity Data to Classify Postural Dysfunction

BACKGROUND

Machine learning (ML) analysis of biometric data in non-controlled environments is underexplored.

OBJECTIVE

To evaluate whether ML analysis of physical activity data can be employed to classify whether individuals have postural dysfunction in middle-aged and older individuals.

METHODS

A 1 week period of physical activity was measured by a waist-worn uni-axial accelerometer during the 2003-2004 National Health and Nutrition Examination Survey sampling period. Features of physical activity along with basic demographic information (42 variables) were paired with ML models to predict the success or failure of a standard 30 s modified Romberg test during which participants had their eyes closed and stood upon a 3-inch compliant surface. Model performance was evaluated by area under the receiver operating characteristic curve (AUC-ROC), balanced accuracy, and F1-score.

RESULTS

The cohort was comprised of 1625 participants ≥40 years (median age 61, IQR 51-71). Approximately half (47%) were diagnosed with postural dysfunction having failed the binarized (pass/fail) scoring mechanism of the modified Romberg exam. Five ML models were trained on the classification task, achieving AUC values ranging from 0.67 to 0.73. The support vector machine (SVM) and a gradient-boosted model, XGBoost, achieved the highest AUC of 0.73 (SD 0.71-0.75). Age was the most important variable for SVM classification, followed by four features that evaluated accelerometer counts at various thresholds, including those delineating total, moderate, and moderate-vigorous activity.

CONCLUSIONS

ML analysis of accelerometer-derived physical activity data to classify postural dysfunction in middle-aged and older individuals is feasible in real-world environments such as the home.

LEVEL OF EVIDENCE

3 Laryngoscope, 133:3529-3533, 2023.

Quantifying defensive behavior and threat response through integrated headstage accelerometry

BACKGROUND

Defensive and threat-related behaviors are common models for aspects of human mental illness.These behaviors are typically quantified by potentially laborious and/or computationally intensive video recording and post hoc analysis. Depending on the analysis method, the resulting measurements can be noisy or inaccurate. Other defensive behaviors, such as suppression of operant reward seeking, require extensive animal pre-training. Inertial tracking and accelerometry can be computationally efficient, but require specialized hardware.

NEW METHOD

We quantified rodent defensive behavior using a commercially available electrophysiology headstage with 3-axis accelerometry integration during a threat conditioning and extinction paradigm. We tested multiple pre-processing and smoothing methods and correlated them against video-derived freezing and suppression of operant bar pressing.

RESULTS

The best approach to tracking defensive behavior from accelerometry was Gaussian filter smoothing of the first derivative. Behavior scores from this method reproduced canonical conditioning and extinction curves. Timepoint-to-timepoint correlations between accelerometry,video, and bar press metrics were statistically significant but modest (largest r = 0.53, between accelerometry and bar press). These increased when traditional thresholding-based analyses were used, at the cost of a loss of temporal resolution (r = 0.97 between thresholded accelerometry and percent time freezing).

COMPARISON WITH EXISTING METHODS

Accelerometry's integration with standard electrophysiology systems and relatively light weight signal processing may make it particularly well suited to detect behavior in resource-constrainedor real-time applications.

CONCLUSIONS

Accelerometry allows researchers already using electrophysiology to assess defensive behaviors without the need for additional behavioral measures or video. The modest correlations between metrics suggest that each measures a distinct aspect of defensive behavior. Accelerometry is a viable alternative to current defensive measurements, and its non-overlap with other metrics may allow for more sophisticated dissection of threat responses.

Naturalistic neuroscience and virtual reality

Virtual reality (VR) is one of the techniques that became particularly popular in neuroscience over the past few decades. VR experiments feature a closed-loop between sensory stimulation and behavior. Participants interact with the stimuli and not just passively perceive them. Several senses can be stimulated at once, large-scale environments can be simulated as well as social interactions. All of this makes VR experiences more natural than those in traditional lab paradigms. Compared to the situation in field research, a VR simulation is highly controllable and reproducible, as required of a laboratory technique used in the search for neural correlates of perception and behavior. VR is therefore considered a middle ground between ecological validity and experimental control. In this review, I explore the potential of VR in eliciting naturalistic perception and behavior in humans and non-human animals. In this context, I give an overview of recent virtual reality approaches used in neuroscientific research.