Using Interpretable Machine Learning for Analyzing Acoustic Startle Response in Rodents from Acoustic Emissions Data

Abstract

Rodent behavior studies in neuroscience and psychology investigate various sensory and motor responses. One such study examines the acoustic startle response (ASR), an unconditional reflex to an unexpected and intense acoustic stimulus characterized by a rapid contraction of facial and skeletal muscles. These contractions generate subtle mechanical forces that propagate through the surface that the animal is in contact with, resulting in acoustic emission (AE) signals. The AE signals travel through the plate as guided ultrasonic waves, carrying information about the animal's motor response to ASR and its physiology, behavior, and underlying psychological state. Various ASR experiments were conducted to assess animal’s reactions to sudden auditory stimuli. The experiments involved three strains of mice (129S6/SvEv, CBA/CaJ, and C57BL/6J) in an open field equipped with AE sensors. The AE data were collected as discrete wavepackets (AE hits) using an amplitude threshold-based acquisition system, with each hit containing signal features that characterize the corresponding movement or behavior. We propose the use of the Explainable Boosting Machine (EBM) to analyze AE data generated by the animal during movement. Our goal is to distinguish ASR-related AE hits from normal AE activity, aiming to analyze the characteristics of each hit to confirm its association with ASR events. This classification is crucial not just for the detection of ASR occurrences but for analyzing the intensity and variability of startle responses across different conditions and strains. The dataset consists of AE-derived features (e.g., amplitude and frequency) combined with biometric factors (e.g., weight, sex) and environmental variables (e.g., lighting conditions, stress exposure). By leveraging EBM’s interpretability, this study identifies key factors influencing ASR and provides a transparent, data-driven approach for analyzing ASR in rodents. EBM provides an effective means to understand complex animal-structure interaction and to analyze the dynamic interplay between body movements and surrounding structures, revealing their mutual influences and shedding light on intricate biomechanical patterns.