Quality control (QC) reports

Kernel Flow provides four quality control (QC) reports to help you evaluate the results of your Flow data:

• NIRS Basic QC
• NIRS Expert QC
• EEG QC
• Sync Accessory Box QC

To view or download QC files:

1. In the Portal, navigate to the dataset you want to analyze:
2. Click Pipelines to open the Pipelines tab. 
3. Click the QC header to expand that section.
4. Click one of the buttons to download the QC file(s).
   Descriptions of the QC reports are listed below.

NIRS Basic quality control report

This report contains a Session Summary and 7 sections, each of which assigns a status to a certain aspect of data quality. The statuses are determined by color and are listed below. 

• Red: indicates an issue
• Orange: indicates a warning or potential issues
• Green: indicates no issues

Sections that are assigned orange or red statuses should serve as flags to improve data recording conditions or to retake the dataset. 

1. Event data: This section provides an overview of Task events received for this dataset. It checks for the presence of a start_experiment event, end_experiment event, and response events (key presses made by the participant). It also lists the number of total events in the session. An orange or red status indicates that a certain type of events were not received. You may ignore this warning if the dataset was not supposed to have Task events or you did not expect certain events, like user responses.
2. Laser Intensity: This section checks whether the target number of photon counts was achieved on channels with a long source-detector separation (SDS). An orange or red status might indicate hardware issues or failure to tune lasers, which can result in detector saturation.
3. Dropped packets: This section checks for dropped data due to USB transfer or other issues. An orange or red status might indicate software issues.
4. Histogram baseline: This section checks for the presence of bad detectors, which usually have high baseline in their histograms. An orange or red status might indicate hardware issues.
5. Saturation: This section checks for saturation of data in channels with short SDS. Saturation may mean that lasers were not properly tuned prior to recording. An orange or red status might indicate hardware issues or failure to tune lasers.
6. Motion: This section checks for the presence of motion in the data. An orange or red status indicates that a participant moved too much, which introduced artifact into the data. Attempt to reduce participant motion.
7. Signal: This section gives a signal strength "score," based on the percent of channels that are analyzable. An orange or red status indicates signal percentage below 40%. Attempt to obtain better signal next time.

NIRS Expert quality control report

This report contains five sections. Click the name in the header of the report to jump directly to that section. 

• Grayplot
• Motion
• Stacked Plot
• Total counts Time Series
• Total counts Topoplot
• Physiology (Physio)
• Retained Channels Topoplot
• Flagged Physiology Topoplot

Grayplot

The grayplot (or carpet plot) visualizes global variations in signal intensity in the raw data. To obtain these visualizations, signal fluctuations are normalized (using robust scaling) and depicted with a grayscale color map:

• Signal above the median manifests as lighter regions on the plot.
• Signal below the median corresponds to darker regions.

The top section displays total photon counts recorded from intra-module channels (source and detector in the same module). Data are ordered by module, and color-coded according to the layout overview on the right. 

The bottom section displays total photon counts from cross-module channels (source and detector from different modules). This section is color coded based on the module to which the detector belongs.

Interpretation

This visualization highlights global artifacts. Vertical bands or artifacts that span all subplots are indicative of global signal variations, which may correspond to physiological noise, motion, or other non-neural sources (brain activations tend to be somewhat more localized). Artifacts that align with large signal fluctuations in the gyroscopes are likely to be motion-related. Global oscillations that align with the task structure may reflect systemic physiology.

Motion

The motion plot visualizes participant head motion during a recording, which introduces noticeable spikes that affect multiple channels simultaneously. For each time point, the number of channels with motion artifacts (spikes) is counted. If more than 2% of channels are affected, that time point is marked as having a motion artifact. The middle plot shows the time course of these artifacts, with time on the x-axis and the percentage of affected channels on the y-axis. The dashed horizontal line represents the 2% threshold. This plot is overlaid on colored blocks indicating the task structure, if applicable.

To summarize, we calculate the fraction of time points with motion artifacts (>2% of channels affected). This fraction is compared to a range observed in Kernel's reliability study (Dubois et al., 2024). The percentile in which the current dataset falls is converted to a score from 1 to 10, with 10 indicating the most motion. This score and its percentile are displayed to the left of the time series plot. On the right side, a spatial map shows the percentage of the recording affected by motion per each specific module. For instance, uniform global coloration indicates global head movements, whereas isolated movement in specific regions (e.g., raising eyebrows) would show as localized dark pink or red color.

Interpretation

The motion plot is useful for identifying and assessing the severity of global or region-specific movement artifacts. Kernel's processing pipelines include motion correction, often allowing for the analysis of data with motion spikes. Still, the motion plot provides valuable insights, indicating whether participants need reminders to remain still or if the setup is causing participant discomfort.

Stacked plot

The stacked plot is very similar to the grayplot with an identical layout. The main difference is that the data fluctuations are shown as lines rather than as a heatmap. Also, the data is averaged for all detectors within a module to make the visualization less crowded. For within module channels, signals from the six detectors are averaged together; for across module channels, signals from the six detectors and all incoming sources are averaged together.

Interpretation

Like the grayplot, the stacked plot is useful in identifying global physiological artifacts, movement artifacts, and other non-neural sources. Spike-like events are somewhat easier to identify in this visualization than in the grayplot, making it a good complement.

Total counts time series

In these plots, total counts for each channel are displayed over the course of the recording. The plot on the left contains within-module channels. The plot on the right contains only across-module channels.

The wavelengths are represented by color: 690nm in cyan and 905nm in magenta. The dashed red horizontal line represents the point of saturation. Saturation may occur if laser power has not been properly adjusted prior to recording (see Tuning the lasers).

Interpretation

This plot provides another method to detect the presence of global artifacts in the data, e.g. large spikes or baseline shifts affecting several channels. This plot is also useful in detecting dead channels or those with poor signal strength, which will show up as lines with consistent low amplitude.

Total Counts topoplot

In this figure, mean total counts for all within-module channels are represented spatially as a topoplot. The 690nm wavelength is on the left, and 905nm is on the right. The color map is on a log scale to visualize the range of total counts which may span several orders of magnitude.

Total counts topoplot Interpretation

Total counts are a simple proxy for signal strength. This is similar to the live display in the Kernel Flow Desktop Application, however in this static image the data are averaged across the entire recording.

Physiology (Physio)

These figures display the physiology metrics pulled from NIRS data. The Flow2 device has one source in the frontal plate which fires at double the rate of all others, which allows us to sample fast enough to capture even the fastest heart rate. On the leftmost graph, each dot represents a channel formed with this source. The x-axis shows the source-detector distance for each channel. The y-axis shows the Scalp Coupling Index (SCI), which measures the correlation between the intensities recorded for the two wavelengths used in our system, after filtering in the heart rate band (0.5 - 2.5 Hz). If it is high—above the green dashed line at 0.75—it is likely that there is a strong heart rate signal. On the middle graph, data from the 5 channels with the highest SCI are averaged to create a time by heart rate (in beats per minute) line chart, overlaid on the color-blocked task event sequence (if applicable). On the rightmost graph, the same data is converted to the frequency domain to create a power spectrum. The overlaid red line is fit to the peak of the heart rate based on the black line. The dashed blue line represents the fit to the "aperiodic" component of the power spectrum.

Interpretation

The physiology plots are useful for analyzing data on heart rate during a data recording. If the SCI of the selected channels is low, or the time course of the heart rate is very noisy (large jumps), it may indicate that there was poor contact between the fast firing source and the forehead. Physiological metrics derived from this data may be of poor quality.

Retained Channels topoplot

Flow2 Retained Channels Example

 

DevKit Retained Channels Example

 

This visualization shows the channels that will be retained for analysis purposes in five different ranges of source-detector separations (SDS). SDS refers to the distance between the light source and the detector for a given channel. The decision for whether a channel is retained takes into account total counts and peak counts, as well as the shape and height of the histograms at the detector.

In the figures, the presence of color indicates a channel whose data will be retained for analysis.In the top figure,all SDS ranges are overlapped on a topoplot. In the five plots in the middle of the visualization, the SDS ranges have been separated by range and color.

In the bottom of the visualization, the figures are further quantified by expressing the exact number of channels retained for analysis in each SDS, separated by brain region. 

Interpretation

Retained channels are a proxy for how much analyzable data was obtained from this participant in this dataset, based on the strength of the signal and the presence of artifacts. You should expect to see higher numbers (and more filled-in topoplots) for the lower ranges of SDS since there is less distance for the light to travel and therefore less risk of the light being blocked in its path. The most important SDS for analysis of NIRS data are 15-25mm and 25-35mm. Higher SDS may not have many retained channels.

[Back to top of page]

Flagged Physiology Topoplot

Depending on a participant's head shape, certain sources, detectors, or modules may not fully contact the head, floating millimeters above the surface. In such cases, you may see high signal strength reported in the Flow UI's module map and most channels retained in the retained channels topoplot, yet expected neural responses might not be observed. The flagged physiology topoplot complements the total counts and retained channels topoplots by highlighting such issues.

We first calculate the "ground truth" heart rate by filtering all retained within-module channels (that have <20 mm SDS) in the heart rate frequency band (0.4-2 Hz), averaging the channels, and creating a power spectrum for the average. The ground truth heart rate is identified as the frequency at which the power spectrum peaks. We then measure how much the average heart rate for each within-module channel deviates from this ground truth. If a channel's heart rate deviates by more than 10% in either direction, it is colored dark red in the topoplot.

Interpretation

The flagged physiology topoplot helps evaluate the likelihood of detecting neural activation in regions with high signal strength and retained channels. For instance, when hoping to record a strong finger-tapping response, ensure the participant exhibits high signal strength in the motor cortex as indicated by the UI, has retained channels in longer SDS in the retained channels topoplot, and is colored light pink in the flagged physiology topoplot.

EEG quality control report

This report contain four graphs:

• Time course of voltage
• Power spectral density
• Coupling
• Q-Q Plot (normal distribution test)

Time course of voltage

Upper Left. This plot visualizes the EEG voltage time course across the entire recording for each channel, color coded according to the legend on the right. The time courses are stacked along the y-axis. If there is any missing data, it is visualized as a fraction of missing data samples per time point (y-axis on the right), and plotted in black.

Interpretation

This plot is useful in determining noisy channels and assessing overall quality of the EEG signal, including if any data are missing.

Power spectral density (PSD) for each channel

Upper Right. This power spectral density (PSD) plot captures the distribution of EEG signal power across frequencies from 2 to 40hz (the typical band considered in EEG), for each channel.

Interpretation

The power spectrum of EEG data typically falls with a 1/f slope (the aperiodic component). Some peaks are expected in specific frequency bands, the most prominent of which is typically centered around 10hz (“alpha” rhythm, especially prominent on occipital electrodes). Spikes in the power spectrum may be indicative of noise sources.

Coupling

Lower Left & Center. A (proprietary) coupling index is computed for each electrode, which varies between 0 and 1, where 1 is the best signal. This index is a proxy for how well "coupled" an electrode is to the scalp. The plot on the left shows the time course of coupling for each electrode (aligned with the time course of voltages), while the plot to the right shows the distribution of coupling values over time for each channel.

Interpretation

Electrodes with better signal will yield less noisy signals, so higher signal strength (>=0.85) is desirable. 

Q-Q plot (test of normal distribution)

Lower Right. The distribution of the values of each channel is visualized against a normal distribution, in a plot known as a Q-Q plot (quantile-quantile).

Interpretation

If values are normally distributed (desirable), they will lie on a straight line (the 45 degree dashed line). Deviations from normality are indicative of artifacts (heavy tails, etc.).

[Back to top of page]

Sync Accessory Box quality control report

The Sync Accessory Box QC is only relevant if you purchased a Sync Accessory Box and it was properly plugged into the data acquisition computer during the recording of this dataset.

This single multipanel figure displays the outputs from each of the Sync Accessory Box streams as a time series over the course of the experiment. In the analog channels, the data are visualized with a continuous y-axis (voltage). In the digital and comparator channels, a logical “high/low” is used to visualize the data.