Optimizing Data Quality in Real-World Scenarios
The best way to ensure high-quality data is to prioritize accuracy during the recording process. Capturing clean, reliable data from the start reduces the need for corrections later. For tips on improving test quality during data collection, please see our related articles.
That said, real-world testing conditions aren’t always perfect. Sometimes, data may be recorded under less-than-ideal circumstances. In these cases, post hoc analysis and filtering can help clarify the data and make it more useful for interpretation.
This article will help you determine whether your existing recording is high quality as-is, or if it requires adjustments. We’ll guide you through deciding whether the data you have is already interpretable—or if it can be improved through post-processing.
| Description | Image |
| Normative Data | |
| Internal Variability | |
| Remove 50/60Hz Noise | |
| Remove Biological Noise | |
| Fix Baseline | |
| Remove Outliers | |
| Fourier Transform |
Standard Deviation
Results can either be compared to reference data and shown to reflect signal variability within a recording.
Standard deviation from normal reference data
Standard tests can be interpreted with the help of reference, or normative, data. Reference data is represented in both the graphs and marker tables. The default number of standard deviations (SD) is set to 2 and may be changed in Configure System.
Reference data is overlaid onto the graphs and represented in the marker table. Notice in the graph below how each marker has a box around it representing the normal range of each marker. Additionally, reference data is shown in the marker tables; here normal ranges are color-coded as red/yellow/green. Green is within 1 SD, yellow is within 2 SD, and red is outside of 2 SD.
To display normative data in this way, navigate to the pop-out menu by clicking on Display. In the display tab, look for "graph label options" at the bottom, and select Norms.
Standard deviation from internal variability
Version 7 software introduces a new way to identify test variability. This method works by calculating the standard deviation from each data point. Think of this as a deviation from the mean. This view shows you the internal variability of all recorded sweeps. Therefore, if you record three sweeps that are all very similar, the normative range will look fairly narrow. However, recording sweeps that are very different from one another will create a much broader distribution. This variability indicator can be printed with your final waveforms.
You can access this feature from the V7 toolbar by clicking on Display SD[2]. The [2] here represents two standard deviations. If you would like to see either 1 or 3 standard deviations, then change this setting in the setup tab, as described below.
Change the number of standard deviations by clicking into the setup tab, finding the "Result SD" section, and choosing the desired number of standard deviations.
Post Hoc Noise Removal
In version 7 software, there are several new ways for you to filter recorded data post hoc. These methods are made to be used after recording is complete. Options for post hoc filtering include:
- Notch removal - Filter out a specific frequency artifact such as 50 or 60 Hz line noise
- Bandpass filter - Adjust the high and low bandpass to filter out biological noise
- Toggle trend removal - Reduce any upward or downward trends associated with eye drift during testing
In the new Espion version 7 toolbar, you will see the following options. Notice that you always have the ability to overlay original and reset original waveforms, so feel free to make changes.
Erase specific frequency noise with a notch filter
Electrical (50 or 60 Hz) noise can create artifacts in your ERG recording. Be mindful when setting up the testing space and utilize runtime filters. However, if you still find electrical noise in your recording, apply the notch filter.
Note: Confirm a noise frequency with a simple calculation. (Number of spikes) * (milliseconds) * (multiplier to convert milliseconds into seconds) In the example below, we count 15 spikes over the course of 250 milliseconds, so we multiply (15 spikes) * (250 milliseconds) * (4 to make one second) = 60 spikes per second, or 60 Hz.
Image: Standard ERG test before (left) and after (right) a 60Hz notch filter is applied.
Apply the notch filter by selecting remove noise from the V7 toolbar.
You may check or change the notch filter frequency in the "analysis" tab of the pop-out menu.
Remove biological noise with a bandpass filter
Bandpass filters are applied to every step of a test in order to isolate the anticipated range of cellular responses. A bandpass filter can be applied to noisy recordings in cases of biological noise interference from patient's eye or face movements.
Image: Standard ERG test before (left) and after (right) the low cutoff filter is changed.
Apply the bandpass filter by selecting band filter from the V7 toolbar.
To adjust the bandpass filter settings, go to the analysis tab of the pop-out menu. Choose from one of the options below:
- Manual - manually adjust the values
- Band - smoothing band from 0.625 - 30 Hz
- OPs - oscillatory potential bandpass range 75 - 300 Hz
- ISCEV - ISCEV standard 0.1 - 300 Hz
- Pupil - used for pupillometry 0 - 1 Hz
Watch this video to see an example of post-hoc bandpass filter application.
Fix baseline drift with trend removal
It is not uncommon for the patient's eyes to drift slightly during ERG recording. This eye movement creates a noticeable upward or downward skew in the graph. Trend removal is designed for use in situations where a graph should start and end at zero. It works by calculating the slope of the graph and repositioning it around zero.
Image: Standard ERG test before (left) and after (right) the baseline has be corrected with trend removal.
To use trend removal, toggle the Toggle Trend button in the V7 toolbar. This option is also available in the "results" section of the pop-up window. You may choose to apply trend removal to the average result or to individual waveforms.
Remove outlying sweeps
When a few aberrant sweeps affect the average, outliers can be removed through manual sweep selection or root, mean, square (RMS) sweep removal method.
Manual Sweep Removal
In the "results" tab, select show all to see all sweeps. In cases where the data is widely dispersed, change the scaling to better see the sweeps; a scale of x10 can be useful. From here, double click on a sweep to remove it. All sweeps can be restored again by clicking on Enable All.
Root, Mean, Square (RMS) Sweep Removal
The root, mean, square (RMS) technique is best used with very large data sets. It works by first squaring each data point to eliminate negative numbers, finding the mean of all data points, then taking the square root. By doing so, we turn each complex waveform into a single number. When this is done for each sweep, we can begin to use the software to distinguish outliers using mathematical comparisons. To start, it can be useful to scale traces from the "results" tab. Then navigate to the "advanced" tab to select your sweep removal method.
- Remove sweeps by Amplitude - enter the minimum and maximum amplitudes to include
- Remove sweeps by RMS Statistics - set a percentage (%) of total outliers to remove (both highest and lowest values)
- Remove sweeps by Largest RMS - set a percentage (%), remove a % of the largest outliers
- Remove sweeps by Smallest RMS - set a percentage (%), remove a % of the smallest outliers
Click on restore all to undo any changes.
Fourier Transform (FFT)
The Fourier transform is a powerful tool for breaking a signal down into its fundamental components. Just as reading a recipe reveals the ingredients that make up a dish, the Fourier transform reveals the frequencies that combine to form a given waveform. By applying the fast Fourier transform (FFT), we can isolate and measure the specific frequency components within a signal—turning complex waveforms into clear, quantifiable data.
Image: A 30Hz ERG response (left) with the Fourier transform (right) that indicates 30Hz response.
To show the Fourier transform, press the "Display FFT" button in the toolbar.
Comments
0 comments
Article is closed for comments.