Objective information-theoretic algorithm for detecting brainstem-evoked responses to complex stimuli.

Area of Science:

• Auditory Neuroscience
• Neurophysiology
• Signal Processing

Background:

• The frequency-following response (FFR) is an auditory-evoked potential reflecting brainstem encoding of complex sounds like speech.
• Current FFR detection relies on subjective interpretation, limiting its clinical and empirical utility.
• Objective FFR analysis is crucial for understanding auditory processing, speech disorders, and brain plasticity.

Purpose of the Study:

• To develop and validate an objective, automated procedure for detecting FFRs elicited by complex auditory stimuli, including speech.
• To quantify the shared information between acoustic stimuli and neural FFRs using mutual information (MI).

Main Methods:

• Mutual information (MI) was computed between spectrographic representations of FFRs and their evoking stimuli.
• A computational model simulated FFRs to establish a criterion threshold (θMI) for response detection at a +3 dB SNR.
• The MI metric was applied as a binary classifier to human FFR recordings (n=35) and sham recordings to assess performance.

Main Results:

• The MI metric achieved 93% overall accuracy in distinguishing true FFRs from sham recordings, outperforming human observers (∼75% accuracy).
• Receiver operating characteristic analysis showed high sensitivity (97%) and specificity (85%) for the MI metric.
• MI demonstrated a monotonic increase with signal averaging trials, suggesting its utility as a stopping criterion for signal averaging.

Conclusions:

• Mutual information provides a robust and completely objective method for automated FFR detection.
• This objective approach can enhance the reliability of evoked potential speech audiometry, minimizing subjective interpretation and human error.
• The MI metric offers a quantitative tool for evaluating speech-evoked brainstem responses in clinical and research settings.