Universal automated classification of the acoustic startle reflex using machine learning.

The startle reflex (SR), a robust, motor response elicited by an intense auditory, visual, or somatosensory stimulus has been widely used as a tool to assess psychophysiology in humans and animals for almost a century in diverse fields such as schizophrenia, bipolar disorder, hearing loss, and tinnitus. Previously, SR waveforms have been ignored, or assessed with basic statistical techniques and/or simple template matching paradigms. This has led to considerable variability in SR studies from different laboratories, and species. In an effort to standardize SR assessment methods, we developed a machine learning algorithm and workflow to automatically classify SR waveforms in virtually any animal model including mice, rats, guinea pigs, and gerbils obtained with various paradigms and modalities from several laboratories. The universal features common to SR waveforms of various species and paradigms are examined and discussed in the context of each animal model. The procedure describes common results using the SR across species and how to fully implement the open-source R implementation. Since SR is widely used to investigate toxicological or pharmaceutical efficacy, a detailed and universal SR waveform classification protocol should be developed to aid in standardizing SR assessment procedures across different laboratories and species. This machine learning-based method will improve data reliability and translatability between labs that use the startle reflex paradigm.

Automated classification of acoustic startle reflex waveforms in young CBA/CaJ mice using machine learning.

BACKGROUND: The acoustic startle response (ASR) is a simple reflex that results in a whole body motor response after animals hear a brief loud sound and is used as a multisensory tool across many disciplines. Unfortunately, a method of how to record, process, and analyze ASRs has yet to be standardized, leading to high variability in the collection, analysis, and interpretation of ASRs within and between laboratories. NEW METHOD: ASR waveforms collected from young adult CBA/CaJ mice were normalized with features extracted from the waveform, the resulting power spectral density estimates, and the continuous wavelet transforms. The features were then partitioned into training and test/validation sets. Machine learning methods from different families of algorithms were used to combine startle-related features into robust predictive models to predict whether an ASR waveform is a startle or non-startle. RESULTS: An ensemble of several machine learning models resulted in an extremely robust model to predict whether an ASR waveform is a startle or non-startle with a mean ROC of 0.9779, training accuracy of 0.9993, and testing accuracy of 0.9301. COMPARISON WITH EXISTING METHODS: ASR waveforms analyzed using the threshold and RMS techniques resulted in over 80% of accepted startles actually being non-startles when manually classified versus 2.2% for the machine learning method, resulting in statistically significant differences in ASR metrics (such as startle amplitude and pre-pulse inhibition) between classification methods. CONCLUSIONS: The machine learning approach presented in this paper can be adapted to nearly any ASR paradigm to accurately process, sort, and classify startle responses.

Alterations in peripheral and central components of the auditory brainstem response: a neural assay of tinnitus.

Chronic tinnitus, or "ringing of the ears", affects upwards of 15% of the adult population. Identifying a cost-effective and objective measure of tinnitus is needed due to legal concerns and disability issues, as well as for facilitating the effort to assess neural biomarkers. We developed a modified gap-in-noise (GIN) paradigm to assess tinnitus in mice using the auditory brainstem response (ABR). We then compared the commonly used acoustic startle reflex gap-prepulse inhibition (gap-PPI) and the ABR GIN paradigm in young adult CBA/CaJ mice before and after administrating sodium salicylate (SS), which is known to reliably induce a 16 kHz tinnitus percept in rodents. Post-SS, gap-PPI was significantly reduced at 12 and 16 kHz, consistent with previous studies demonstrating a tinnitus-induced gap-PPI reduction in this frequency range. ABR audiograms indicated thresholds were significantly elevated post-SS, also consistent with previous studies. There was a significant increase in the peak 2 (P2) to peak 1 (P1) and peak 4 (P4) to P1 amplitude ratios in the mid-frequency range, along with decreased latency of P4 at higher intensities. For the ABR GIN, peak amplitudes of the response to the second noise burst were calculated as a percentage of the first noise burst response amplitudes to quantify neural gap processing. A significant decrease in this ratio (i.e. recovery) was seen only at 16 kHz for P1, indicating the presence of tinnitus near this frequency. Thus, this study demonstrates that GIN ABRs can be used as an efficient, non-invasive, and objective method of identifying the approximate pitch and presence of tinnitus in a mouse model. This technique has the potential for application in human subjects and also indicates significant, albeit different, deficits in temporal processing in peripheral and brainstem circuits following drug induced tinnitus.

AMTAS(®): automated method for testing auditory sensitivity: II. air conduction audiograms in children and adults.

OBJECTIVE: This study was designed to evaluate an automated pure-tone audiometric procedure (AMTAS(®)) for 4-8 year-old children, and a quality assessment method (QUALIND(®)) that predicts the accuracy of the test. DESIGN: Children were tested with AMTAS and conventional manual air-conduction audiometry. A group of adults was tested for comparison. STUDY SAMPLE: Eighty-one 4-8 year-old children and 15 adults. Most had normal hearing. RESULTS: For most subjects (93% of adults and 91% of children) differences between AMTAS and manual thresholds were similar to differences that occur when two experienced audiologists test the same subjects. QUALIND detected the inaccurate audiograms with a sensitivity of 71% and a specificity of 91%. When inaccurate audiograms identified by QUALIND are excluded, the accuracy of AMTAS is similar to the accuracy of manual audiometry. CONCLUSIONS: AMTAS produces accurate air-conduction audiograms in a high proportion of 4-8 year-old children and adults. QUALIND successfully identified most inaccurate AMTAS audiograms. The method can decrease the cost and increase efficiency and accessibility of hearing testing.

Neural correlates of age-related declines in frequency selectivity in the auditory midbrain.

Reduced frequency selectivity is associated with an age-related decline in speech recognition in background noise and reverberant environments. To elucidate neural correlates of age-related alteration in frequency selectivity, the present study examined frequency response areas (FRAs) of multi-unit clusters in the inferior colliculus of young, middle-aged, and old CBA/CaJ mice. The FRAs in middle-aged and old mice were found to be broader and more asymmetric in shape. In addition to a decrease of closed/complex FRAs in both middle age and old groups, there was a transient decrease in V-shaped FRAs and a concomitant increase in multipeak FRAs in middle age. Intensity coding was also affected by age, as observed in an increase of monotonic responses in middle-aged and old mice. While a decline in low-level activity began in middle age, reduced driven rates at suprathreshold levels occurred later in old age. Collectively, these results support the view that aging alters frequency selectivity by widening excitatory FRAs and that these changes begin to appear in middle age.

Timing is everything: temporal processing deficits in the aged auditory brainstem.

This summary article reviews the literature on neural correlates of age-related changes in temporal processing in the auditory brainstem. Two types of temporal processing dimensions are considered, (i) static, which can be measured using a gap detection or forward masking paradigms, and (ii) dynamic, which can be measured using amplitude and frequency modulation. Corresponding data from physiological studies comparing neural responses from young and old animals using acoustic stimuli as silent gaps-in-noise, amplitude modulation, and frequency modulation are considered in relation to speech perception. Evidence from numerous investigations indicates an age-related decline in encoding of temporal sound features which may be a contributing factor to the deficits observed in speech recognition in many elderly listeners.

Impaired gap encoding in aged mouse inferior colliculus at moderate but not high stimulus levels.

Age-related deterioration of auditory temporal acuity has been identified as a contributing factor in presbycusis. In the present study, the effects of aging and stimulus level on gap encoding and gap recovery functions were investigated by measuring near-field auditory evoked potentials in the inferior colliculus of eight 3 month old and eight 24 month old CBA/CaJ mice, in response to gap stimuli embedded in broadband noise (40, 60, and 80 dB SPL). Gap encoding was assessed by measuring latencies and amplitudes of peak features of the near-field response, and also with a procedure that calculated the root mean square of the response within specific time windows. The chief differences in gap encoding between young and old mice were longer gap thresholds, slower recovery functions, and longer response peak latencies for old mice at 60, but not 80 dB SPL, although the latency of the earliest measured peak remained delayed for this condition in the old compared with the young mice. These results demonstrate that age-related changes in temporal acuity may interact with stimulus level, and suggest that adequate amplification may be critically important for maintaining temporal acuity with advancing age.

Behavioral and neural measures of auditory temporal acuity in aging humans and mice.

Three experiments compared auditory temporal acuity in humans and in the behavior and single cells in the inferior colliculus (IC) of mice, to establish the comparability of aging effects on temporal acuity across species, and to suggest a neural foundation. The thresholds for silent gaps placed in white noise (MGTs) were similar in young mice and young humans, and increased in some but not all old humans and old mice. Neural MGT in the most sensitive cells of both young and old mice was comparable to behavioral MGT in the young of both species, but older mice had more cells with very high MGT. Human listeners were selected to have minimal absolute hearing loss. Older mice had significant hearing loss that was correlated with MGT in behavioral, but not in neural, measures. Some old mice and some old IC cells, however, had low MGTs coupled with elevated absolute hearing thresholds. Age-related changes in temporal acuity appear comparable in humans and mice. The data suggest a common deficit in neural mechanisms.