Bayesian Weighting
What is Bayesian Weighting?
Bayesian Weighting is a tool designed to assist the clinician when testing in less than optimal conditions. While giving sweeps with less noise a higher “score”, sweeps with more noise are not given the same level of importance in the overall recording. Each sweep is analysed and not simply accepted ot rejected as with traditional averaging but is given a unique signficance based on its level of noise. Baysian Weighting can be used in all typical ABR testing situations and will be effective when EEG levels vary during recording.
When and why to use Bayesian Weighting?
When
- As Bayesian Weighting does not change the response of a waveform, you may use Bayesian in all typical ABR recording situations
Why
- The optimum use of all data reduces test time.
- More stable recordings, as residual noise will never suddenly rise (and a good waveform deteriorate) during recording, even if patient starts to be uneasy.
- The difficulty of selecting the optimum rejection rate is reduced, as a softer rejection rate can be used without the noisier sweeps contaminating the more quite sweeps.
When is Bayesian Weighting less Relevant?
If the patient does not have a fluctuating EEG during the session: Bayesian weights noisy sweeps less and quite sweeps more. In a situation where all sweeps are the same, they will be weighted equally. This is identical to normal averaging. Hence, there is no difference between recordings with and without Bayesian.
If a very tough rejection level is set: In this case, all noisy sweeps are simply rejected for both Bayesian and non-Bayesian recordings. Only the most quite sweeps are accepted, and you have a situation almost similar to above. With softer rejection criteria (e.g. 80μV) you would benefit from the contributions of the more noisy sweeps in your averaging thus arriving at your desired residual noise level in less time.
If you run long enough to get a low residual noise level
If two waveforms each have 40nV residual noise, then they will look equally clean. It does not matter whether you have used 1.000 or 10.000 sweeps to get there or what type of weighting is used. Weighting will simply get you there faster if the patient has fluctuating EEG during the session. It should be noted that two waveforms each with 40nV residual noise may exhibit minor variations in wave morphology due to e.g. the frequency distribution of the residual noise. This may differ with different test situations, and quite patients tend to provide residual noise with more high frequency content and less low frequency content, which may look nicer to the eye. So keeping the patient as relaxed as possible is still recommended.
References
Elberling, C. & Wahlgreen (1985). Estimation of auditory brainstem response, ABR, by means of Bayesian interference. Scand. Audiol (14) 89-96.
