Fig. 2
From: An open-source deep learning-based toolbox for automated auditory brainstem response analyses (ABRA)
Model architecture for the wave 1 peak finding algorithm. The input (a 244-point ABR waveform spanning 10 ms) passes through two layers of convolution, batch normalization, max pooling (after ReLU activation), and dropout. The dimensionality of the output is then reduced through two consecutive fully connected layers using ReLU activation which returns the prediction of the time point of the wave 1 peak.
