Auditory brainstem response images

Contain ABR image render from waveform signal.

All waveforms, regardless of source, were processed through an identical three-step pipeline before rendering. First, a fourth-order Butterworth bandpass filter spanning 30–1500 Hz was applied using zero-phase forward-backward filtering (scipy.signal.sosfiltfilt). Second, each filtered waveform was resampled to a uniform target rate of 20,000 Hz using Fourier-domain resampling (scipy.signal.resample). Third, each resampled waveform was truncated to exactly 300 samples, corresponding to a 15 ms post-stimulus analysis window at 20 kHz, matching the acquisition window of the internal dataset.

Preprocessed traces were stacked vertically using the Matplotlib library, with higher-intensity recordings plotted at the top and lower-intensity recordings at the bottom, matching conventional clinical ABR display. The top intensity was the maximum intensity recorded in the dataset. One image was rendered per subject (Southampton), per subject-frequency pair (PhysioNet, at 1 kHz and 4 kHz separately), or per subject (Mendeley). Output images were 2351 1951 pixels at 100 DPI to match the internal dataset. Subject-specific waveforms were vertically stacked in descending order of stimulus intensity, from the maximum recorded intensity down to 30 dB in 10 dB decrements, and rendered as 2D images. An image was classified as abnormal if wave V was absent in any signal recorded at or above 30 dB. To normalize the waveform amplitudes, a global scaling factor was determined by calculating the peak absolute amplitude (A_max) of the top-most (highest intensity) waveform. Each waveform in the stack was divided by A_max and scaled such that its normalized range fell within a target amplitude spanning a predefined percentage of the canvas height (specifically tested at 10%, 15%, 20%, 25%, and 30%). The vertical spacing between waveform baselines was set to 10% of the canvas height (195.1 pixels). To center the stack of n waveforms vertically, the baseline offset of the lowest-intensity trace was positioned at a margin of (1951 - 195.1(n - 1)) / 2 pixels from the bottom of the canvas, with subsequent baselines offset progressively by 195.1 pixels. The waveforms were plotted using the ax.plot method of the Matplotlib library as solid blue lines (hex color #0000FF) with a linewidth of 2.3 points, using a round cap style (solid_capstyle='round') and with anti-aliasing disabled (antialiased=False).