ROI Activation Feedback Calculation
For many (clinical) neurofeedback applications, it is desirable to visualize the mean activation level from selected ROIs to the participant in the scanner. The Neurofeedback dialog can be used to select one or more ROIs and to prepare feedback display options as described in the previous topic. In this section, details about the calculation of neurofeedback signals from ROIs are described and how calculation is controlled using options in the Feedback settings field in the Neurofeedback dialog.
Given a baseline level bl (see below), the feedback value fb for the current time point (without/prior to averaging) with value val is calculated within a neurofeedback trial simply as follows:
fb = (val - bl) / bl * 100
This results in a percent signal value that is used to fill the thermometer display. To convert this value into a corresponding fill level of the thermometer, the resulting percent signal value is related to the maximum percent signal value as specified in the Max PSC 1 field (the Max PSC 2 field is used for the second thermometer display in case that two ROI feedback thermometers are shown as described earlier). If, for example, the calculated fb value is 2,5% and the max psc value is set to 5%, the thermometer is filled half, i.e. 5 out of 10 rectangles will be filled since fb/max_psc = 0.5. To find good values for the maximum psc, it might be useful to observe achievable neurofeedback amplitudes for specific ROIs in localizer or test runs. It is also possible to automatically set the max psc value by enabling the Set to max measured option, which will set the value in the Max PSC 1 field to the maximum feedback value obtained until the current time point.
The Average feedback values spin box can be used to specify the number of time points that are averaged to calculate the current feedback value. With a value of "1" no averaging takes place, i.e. the feedback value is based directly on the current time point's ROI signal value. If the averaging value is larger than 1, the last N calculated feedback values are averaged to calculate the current feedback value. This effectively implements a temporal low-pass filter (smoothing) that "stabilizes" the feedback signal. In order to avoid, however, additional delays during TR-by-TR feedback, we recommend to use a rather small value for averaging time points in the range of 2 to 5 (default: 3).
Baseline Identification and Calculation
Normally the baseline is calculated based on the "rest" or "fixation" condition prior to an active modulation condition. TBV identifies automatically the first condition in the stimulation protocol specified in the TBV Settings file as the "baseline" condition. If another condition is the "baseline" condition in your protocol, select it in the Condition box in the Feedback baseline data points field (see snapshot below). Note, however, that it is highly recommended to always define a baseline condition as the first condition in a stimulation protocol since the first condition is also identified as the baseline condition in statistical (GLM) calculations.
When calculating the baseline for time points of a neurofeedback block, TBV uses the baseline interval defined just before the interval of the current feedback condition, which can be any other "modulation" condition (e.g. one or more different mental tasks). Using the baseline just before a modulation block is useful since only values in a small window are used which reduces the effect of global drifts on the calculated baseline and feedback values. Because of the sluggishness of the BOLD response, the values defined by a baseline interval should not be used as defined in the protocol but should take the hemodynamic delay into account. In order to protect the BOLD decay from a previous modulation condition, values at the begin of the baseline condition need to be excluded. The number of points to exclude can be specified with the Shift at begin spin box. The default value entered here by TBV is calculated simply by dividing the value 6 by the time-to-repeat value (TR in units of seconds) specified in the TBV settings file; if, for example, the TR is 2000 milliseconds, the default shift at the begin of the previous baseline will be 6/2 = 3 time points. If the previous baseline condition occurred, for example, from time points 60 - 69 (10 data points), the actual values considered for baseline calculation would be from 63 - 69 in case of value 3 for the shift at the begin of the baseline interval. In a similar way one can also extend points "to the right side" at the end of a baseline interval since it takes some time until the BOLD signal rises in the modulation block. Since it is expected that the signal starts to rise already after about 2-3 seconds, the shift into the modulation interval should be, however, smaller than at the begin of the baseline condition. As default, TBV calculates the shift at the end of the previous baseline interval as a third of the shift at the begin; in our example this leads to a value of 1 (3/3). The default value can be changed, however, using the Shift at end spin box. The final points that would be used as default to calculate the baseline for the interval 60 - 69 with a TR of 2 seconds would, thus, be 63 - 70. The signal values of the identified data points (8 in the example) will be averaged to obtain the baseline level bl for the subsequent modulation block (see above). Note that 4 points are minimally required for a pre-modulation baseline (at least 10 are recommended), otherwise no feedback output is produced in the thermometer display.
As an alternative to use the previous baseline for a modulation block, it is also possible to use the baseline value calculated by the GLM analysis of the respective ROI time course data (beta value of the constant predictor, also called "b0"). To enable this feature, turn on the Use ROI-GLM baseline (b0) option. While this option provides a more stable baseline estimation than the rather small amount of values in a pre-feedback baseline interval, the calculation can lead to biased values in case of signal drifts. It is, thus, recommended to use this option only in case that a linear trend confound predictor has been added to the design matrix. The addition of a linear trend predictor (but not higher-order drift corrections that may lead to unstable estimates) is recommended in general, i.e. also for the pre-modulation baseline approach since the time course values used to calculate the baseline and neurofeedback values will be corrected for estimated linear trends in the selected ROI(s). ROI time course detrending is enabled as default (if a linear trend confound predictor is included in the design matrix) but it can also be turned on/off using the D key or the Display Detrended Time Courses item in the View menu.
Note. Since a baseline is not calculated for the first 5 time points of a functional run, a sufficiently long baseline condition should be used at the beginning of a functional scan so that after the first 5 time points enough additional baseline points are available before the first neurofeedback trial starts. This is enforced automatically since TBV v3.2 because feedback values are only calculated if more than 10 data points have been scanned. Note that you may need a few more baseline points at the begin in case that you also skip volumes at the begin of a functional run. It is recommended to specify at least 10 time points (better 20) as the first interval of the baseline condition at the begin of a neurofeedback run.
Notes for previous versions. In previous versions (v3.0 and earlier), TBV identified baseline condition(s) by checking the name of each condition of the current protocol and assumes a baseline condition if the condition name contains the string "fix", "baseline" or "rest". This check is performed case-insensitive, e.g., both "Fixation" and "Cond-rest" would be identified as baseline conditions. Alternatively one can append "_0" to any condition name in the protocol to indicate that the respective condition intervals should be used as baseline epochs. TBV looked for a baseline condition in the past of the current time point within a window maximally of the size specified in the Baseline detection window spin box. The first baseline epoch discovered before an active neurofeedback trial was used for baseline calculation (averaging). If the window is so large that it encompasses earlier baselines, those epochs were not included in the calculation, i.e. just the baseline before an active neurofeedback trial was used. Since the search for a baseline epoch begins at the current time point, the value in the Baseline detection window spin box (default: 60) should be at least as large as the sum of the (longest) baseline and the (longest) active neurofeedback condition. The values of the baseline are not used directly but accommodate for the expected hemodynamic delay that was internally calculated simply as 6s / TR with TR in units of seconds. In order to prevent that values of the early rise of BOLD response in the next feedback trial is included in the baseline calculation, the hemodynamic delay was actually not added to the end of the rest epoch, i.e. the actual values used for baseline calculation of interval 60-70 would have been 63 - 70. Note also that the detected baseline period must have more than 3 data points otherwise the last (previously) calculated baseline is used as a fall-back option.
Copyright © 2014 Rainer Goebel. All rights reserved.