EC209 - Improving Signal-To-Noise Ratio (SNR) for Optimizing Repeatable Auditory Brainstem Responses Aaron Steinman, Ph.D. Director of Research, Vivosonic Inc. aaron.steinman@vivosonic.com 1
Outline Why is Signal to Noise Ratio (SNR) relevant? A look at noise reduction A look at signal enhancement An understanding of what contributes to a better SNR and what that means clinically 2
Why is SNR Relevant? What do you, the clinician, want in ABR? Ability to diagnose quickly and confidently Ability to perform a complete assessment within one test session Confidence that the information can be used for accurate assessment, intervention, monitoring 3
Why is SNR Relevant? Typical pain points (why you can t always get what you want): Noisy / unclear results Cannot get results in certain environments Performance sensitive to skin preparation Poor repeatability Cannot determine threshold Takes too long to get results Necessity to continually press pause/resume during testing 4
Why is SNR Relevant? For example, after averaging, you end up with an ABR signal estimate of 10 μv rms and a noise estimate of 10 μv-rms SNR = 20 log 10 Signal Noise = 20 log 10 10 μv rms 10 μv rms = 0 db If increase signal (i.e. response) by 20% SNR = 20 log 10 If decrease noise by 20% SNR = 20 log 10 12 μv rms 10 μv rms 10 μv rms 8 μv rms = 1. 58 db = 1. 94 db We have chosen to focus on noise reduction 5
Noise Reduction Types of Noise Non-synchronous noise e.g. Muscular noise (EMG), eye movement (EOG) Can eliminate with averaging (can take a long time many sweeps) Synchronous noise e.g. Stimulus artifact, powerline harmonics Hard to eliminate with averaging strategies Other noise might really be signal e.g. PAM or 40 Hz superimposed on ABR waveform they are also AEP 6
Noise Reduction You, the clinician, can help reduce the noise Calm, quiet, still subject (reduce myogenic noise) Test in a quiet environment Both acoustically and electrically quiet Equipment adjustments (e.g. braid the electrode wires) Pick stimulus rates that do not lock into powerline harmonics 7
Noise Reduction But when that doesn t work, what can the manufacturer do for you? Improved design of both hardware and software 8
Noise Reduction Hardware designed to reduce the negative effects of noise by preventing the noise from entering the system Filtering Shielding Limited wire length Battery powered Bluetooth Impedance mismatch 9
Noise Reduction Software designed to reduce the negative effects of noise: Technique Pros Cons Filtering Artifact rejection Kalman-weighted averaging SOAP-Kalman SOAP-Kalman Plus Easily eliminate out of band noise. Simple technique to remove noisy sweeps Reduces corruption from noisy sweeps without rejection Optimized to cancel noise including sinusoidal noise SOAP applied to multiple harmonic bands can remove many harmonics Cannot remove noise with similar frequencies to signal. Have to find correct level, may throw out lots of data Does not remove sinusoidal/ harmonic noise effectively Will only remove one dominant sinusoidal noise frequency Possible slight reduction in signal amplitude 10
Noise Reduction Software provides other tools to help with SNR A,B buffers to visually observe repeatability (Correlation Coefficient) A-B waveform to compare amount of noise to signal (Residual Noise) Number of Noise Adjusted Sweeps (Neq) Total number of stimuli (N) is discounted based on the relative noise to determine Neq Difference between N and Neq is an indication of patient and EM noise 11
Noise Reduction Main waveform Independent A and B traces, to determine repeatability, half the number of sweeps in each average A-B trace; subtract trace B from A to remove the ABR; see residual noise Correlation: A, B Residual Noise: A-B 12
Averaging, AR = 25 μv Noisy Data, N = 4,000 SOAP-Kalman Kalman 35 db nhl, Same Data Noise-Reduction Hardware SOAP-Kalman Plus 13
Noise Reduction Summary of Noise Reduction Improved hardware reduces the amount of noise from entering the system Improved software algorithms intelligently remove noise that is not synchronized with the response Visual and statistical tools enable confident determination of repeatability and residual noise 14
Signal Enhancement ABR is an onset response, it is impacted by: Stimulus rate Electrode montage Stimulus type and windowing Superimposing of other AEP MLR (40 Hz), PAM Stimulus intensity (threshold limit where response is discernable) 15
Signal Enhancement Standard stimuli Click Frequency specific tone-bursts Usually 500 Hz, 1 khz, 2 khz, 4 khz Different windows (type and width) Lots of literature with normative data for normal and hearing impaired populations throughout the age groups 16
Signal Enhancement Alternative stimuli Potential to improve the evoked response amplitude at the same intensity level Increase bandwidth: More cochlear response, less frequency specificity Optimizing delays of each frequency component of stimulus so that each component simultaneously arrives at the place of its maximum response in the cochlea Is there a clinical implication by using these alternative stimuli? 17
Signal Enhancement Normalized 2-1-2 Tone-Burst Spectra 0-5 -10-15 Narrow Spectral Width 500 Hz TB 1 khz TB 2 khz TB 4 khz TB Frequency Spectra - Tone-Bursts Spectral Splatter [db] -20-25 Overlap -30-35 -40-45 -50 10 2 10 3 10 4 Frequency [Hz] 18
Signal Enhancement Normalized 2-1-2 Tone-Burst and Narrow Band Chirp Spectra Frequency Spectra - Tone-Bursts and Narrow Band Chirps 0 [db] -5-10 -15-20 -25-30 -35-40 -45 500 Hz TB 1 khz TB 2 khz TB 4 khz TB 500 Hz NB chirp 1 khz NB chirp 2 khz NB chirp 4 khz NB chirp How much overlap? -50 10 2 10 3 10 4 Frequency [Hz] How strong a filter? 19
Tone Burst NB Chirp 500 Hz 20
Signal Enhancement Summary Chirp stimuli give higher amplitude responses and potentially lower thresholds than the corresponding tone bursts Data was collected using hardware and software algorithms designed for noise reduction Outstanding questions: What is the ideal chirp to use? Bandwidth, delay model, stimulus rate, level specific 21
Signal Enhancement Outstanding questions, cont d: How to interpret the results? Need normative data for the population of interest for normal hearing and hearing impaired populations No verified correction factors in clinical populations Using the same equipment, is a threshold of 20 db nhl for tone burst and for narrow-band chirp equivalent? 22
Signal Enhancement Outstanding questions, cont d: If a NB chirp can get a lower threshold than TB, does that only affect the db ehl to nhl conversion? What is the benefit of achieving a lower threshold if it only impacts that conversion? Is there a time savings if need to test another intensity to get the lower threshold? How does this impact hearing aid fitting? 23
Conclusion Noise reduction helps whether or not a response is present. Eliminating noise is best first step take advantage of knowledge base established for traditional stimuli Use repeatability statistics and A,B buffers to visualize noise Software and hardware solutions to reduce noise 24
Conclusion Possible to enhance signal with chirp stimuli But many outstanding questions Toneburst amplitudes may be clinically adequate when using a noise optimized system Noise and/or Signal = SNR = Better visual waveform 25
Conclusion Points to consider: Are NB chirps responses larger than those of TBs because of synchronized firings or because they have wider energy bands? Is a threshold with a narrow spectra equivalent to a threshold with a wider spectra with respect to hearing aid fitting? If tone burst responses are clear, then does the benefit of a larger response from a chirp stimulus outweigh the associated uncertainties? Should NB chirps partner with tone bursts or replace them when determining frequency specific thresholds? 26
Conclusion A better SNR will help you, the clinician, get better quality waveforms and meet your goals: Ability to diagnose quickly and confidently Ability to perform a complete assessment within one test session Confidence that the information can be used for accurate assessment, intervention, monitoring A better understanding of what contributes to SNR will let you make informed decisions on the reliability and clinical suitability of protocols and corresponding results. 27