Abstract
Purpose:
Sweep VEP (visual evoked potentials) is known to be a valuable exam to estimate visual acuity (VA) in infants because of its objectivity and its rapidity. Nevertheless, several studies have pointed out the variability of the responses and sometimes attenuations at some spatial frequencies called “notches”. In a previous study, we demonstrated a relationship between the phase of the signal and these attenuations when the moving-window discrete Fourier transform (MWDFT) was used. Here we evaluated a new method of signal processing which takes in account the phase shifting, based on discrete wavelet transform (DWT) for the denoising and on extended Kalman filtering (EKF) for the estimation of the VA.
Methods:
3 adults with normal VA took part in the study (mean age: 21 years). For each subject, 5 different Bangerter penalties were tested in order to simulate low vision. ETDRS acuity was tested for all the penalties and without the penalty. A checkerboard was presented on a MonPackOne (METROVISION) with a reversal rate of 12 Hz. The sweeps lasted 11 seconds and 5 sweeps were performed for each exam. DWT was performed with the symmetric wavelet “symlet 13”. In this study, artefacts are considered to be high rapid changes of the amplitude of the signal which are mainly due to ocular movements and blinking. The noise, before and after the denoising, was calculated with four adjacent frequencies (10, 11, 13 and 14 Hz). EKF was used to estimate the amplitude of the response. Its performance was compared to that of the conventional MWDFT. After processing, the mean of at least 3 sweeps was calculated for both techniques. VA was estimated with an extrapolation of the regression line to the zero amplitude.
Results:
DWT rejects artefacts as estimated with the standard deviation of the signals which is reduced in average by 35.4% (CI [15.6, 55.2], p=0.05). The signal to noise ratio is improved by 0.25 dB (CI [-1.9, 2.4], p=0.05). These results show that wavelet denoising rejects artefacts without impacting the signal to noise ratio. The RMS value of the difference between ETDRS VA and that of the EKF is 0.28 LogMAR. It is 2.5 times lower than the RMS value obtained with the MWDFT (0.63 LogMAR). EKF VA is closer to ETDRS VA than MWDFT VA (independent T-test gives p=9.45 10-10), particularly for high spatial frequencies.
Conclusions:
The VA given by the new technique is more accurate than that of the windowed Fourier transform.