3.1. Phantom measurements
The dynamic phantom setup we developed was designed for automated system characterization. Given the 32 sources by 32 detectors, manual characterization of the individual channels can become time consuming. Our dynamic phantom automates characterization of all source and detector channels simultaneously. The measurements on the dynamic phantom allowed us to characterize noise equivalent power (NEP) and dynamic range of the detector channels in the CW5 system. NEP corresponds to the amount of light at the detector that has a signal-to-noise ratio of unity. We calculated an NEP of 0.05 pW per root Hz, which is comparable to the NEP of the APDs. This means that the circuitry following the APDs preserves the signal-to-noise ratio. The dynamic range of the instrument is defined by the ratio between the minimum light that saturates the detector and NEP. For the different detectors in the CW5 system the dynamic range varied between 60–70 db with less than 2 % deviation from a linear least squares fit.
The results of the ink titration for four inline source-detectors pairs are shown in . The detected light decreases exponentially with the increase in the absorbance of the solution. NEP and dynamic range are calculated by these curves for all the other source-detector combinations. The increase in noise with increasing signal for the same detectors is plotted in at a 50 Hz bandwidth.
Fig. 5 (a) Measurement of the dynamic range of the system. As a representative example, we show here the power recorded by 4 adjacent detectors (each color is a different source-detector pair) as the absorption of the medium was increased. (b) Measurement of (more ...)
The system was found to have a signal drift less than 0.5% over five minutes.
3.2. Human subjects - baseline measurements
Optical data obtained from the head during rest typically contains a wide variety of oscillations due to systemic and local hemodynamic changes. The major contributions to the optical signal are from cardiac pulsation, respiration, heart rate and blood pressure changes. The average cardiac pulsation frequency of our subjects was 1 Hz (±0.1). In the same subjects, we measured average heart rate oscillations of 0.1 Hz (±0.03), and slightly lower blood pressure oscillations of 0.08Hz. The average respiration frequency was 0.23 Hz (±0.07). Four subjects breathed at a frequency lower than 0.2 Hz and for them we observed frequency- and phase-locking of blood pressure and heart rate oscillations with respiration. The remaining seven subjects had an average respiration frequency higher than 0.27 Hz and the heart rate and blood pressure oscillations were clearly separated and at a much lower frequency than the respiration oscillations.
We cross-correlated the optical data with the systemic signals during our 5-minute baseline runs in the 11 subjects. The mean cross-correlation values of the optical signal over 50 positions on the head, with cardiac pulsation, respiration, blood pressure and heart rate averaged over the 11 subjects and 2 baseline runs, are reported in . The highest cross-correlation value is between optical data and cardiac pulsation; the lowest between optical data and blood pressure. The cross-correlation values are slightly higher at 830 nm than at 690 nm, which agrees with the fact that these oscillations are mostly in the arterial compartment and that oxygenated blood has higher absorption at 830 nm than 690 nm.
Temporal cross-correlation coefficients at zero time delay averaged over the 11 subjects during baseline. The Errors are the standard deviations
The mean cross-correlations values over all the channels are considerably smaller than the maximum values because of the large spatial differences of the amplitudes of these optical signals.
In the slow-breathing subjects, interference between respiration and heart rate/blood pressure oscillations was also observed in the optical data. shows the cross-correlation between optical signal and auxiliaries in representative cases of fast- and slow-breathing subjects. The fast-breathing subject () breathed at a rate of ~15 breaths per minute (0.25 Hz, one breath every 4 sec), the cross-correlation between optical data and respiration is strong in most of the channels. Blood pressure oscillations in this subject have a period of about 0.075 Hz and heart rate oscillations have a period of 0.1 Hz. The cross-correlation between optical data and blood pressure/heart rate oscillations shows the same periodicity and has a lower amplitude than the cross-correlation with respiration. It is interesting to note that different channels show a different phase lag between optical data and blood pressure/heart rate. The slow-breathing subject ) breathed at a rate of 8 breaths per minute (0.13Hz, one breath every 7.5 sec) and there is a locking of blood pressure/heart rate with respiration. In this case the optical signal in the head also shows this locking, and the cross-correlation of the optical data with respiration, blood pressure or heart rate oscillations is temporally identical.
Fig. 6 Time traces of the temporal cross-correlation of all the optical data channels with the auxiliaries in two subjects: (a) and (c) subject breathing fast, (b) and (d) subject breathing slow. (a) and (b) show the cross-correlation for temporal lags spanning (more ...)
Slow breathing tends to exaggerate the parasympathetic feedback to the heart’s pacemaker cells and to the arterial baroreflex. This tendency of the autonomic nervous system to increase parasympathetic feedback during slow respiration could originate in the ventrolateral aspect of the medulla oblongata (VLM)where the integration of afferent inputs to the respiratory and cardiovascular controllers may create the frequency locking with blood pressure and heart rate variations [14
]. This frequency locking is an artifact of the blood pressure regulatory system. Its presence means that the subject is probably relaxed and has a well-functioning autonomic nervous system. The lower natural respiration rate of some subjects reflects normal inter-subject variation in the respiratory set point.
The different temporal delays between optical data and blood pressure/heart rate in different channels, seen in , are consistent among subjects and have a well-defined spatial pattern. Large variance in the phase of cerebral blood flow velocity relative to arterial blood pressure has been observed with transcranial Doppler ultrasonography comparing the middle and posterior cerebral arteries [15
] These phase variations may be attributable to regional differences in cerebral autoregulation as observed in positron emission tomography studies of cerebral blood flow [16
]. The spatial maps of cross-correlation between optical signal and blood pressure/heart rate reveal a delay between the anterior, posterior, and middle areas of the head. shows the spatial maps of cross-correlation between optical data at 830 nm and blood pressure in the same two subjects as shown in . In this figure we show 16–18 frames during one period of the blood pressure oscillation, one image every second for the fast breather and one image every half second for the slow breather, since the blood pressure periods of the two subjects are about 15 and 8 seconds, respectively. In the images the top corresponds to the frontal area, the bottom to the back of the head, and the left and right to the left and right head hemispheres, respectively. The “blood pressure wave” starts anteriorly and propagates to the back and then to the middle of the head periodically. We observed this propagation pattern for blood pressure and heart rate oscillations in all of the subjects. For the cardiac pulsation we didn’t observe any spatiotemporal delay between different positions in the head. In other words, at the cardiac pulsation frequency, every position on the head oscillates at the same time. For the respiration oscillations, apart from the four slow-breathing subjects, in whom respiration propagation through the head followed the blood pressure and heart rate patterns, the spatial-temporal non-uniformity was smaller or not visible. When visible (3 of 7 subjects), the pattern was the same as for blood pressure.
Fig. 7 Maps of the cross-correlation of the optical data with the blood pressure in the same two subjects as in . (a) This subject had a slow respiration period of 15 sec. One image every 1 sec is displayed with the color scale ranged linearly from −0.18 (more ...)
The spatiotemporal structure of this apparent blood pressure wave must arise from the anatomical structure and dynamic properties of the scalp and cerebral vasculature but we are not in a position to explain this observation in more than general terms. The primary reason is that these images are simply the measurements projected onto the head surface. There is no differentiation in this data between variations that come from the scalp and the brain. It is also not possible to differentiate the effects of these two anatomical regions without denser spatial coverage or significant prior knowledge of the anatomy and physiology. The structure of blood vessels is spatially inhomogeneous and this will need to be considered. Cerebral autoregulation may also play a significant role because it introduces frequency dependent phase shifts in the respiratory-related blood pressure variations. Spatial variation in the skin and skull thickness together with cerebral autoregulation effects could explain the observed blood pressure waves.
3.3. Human subjects - paced breathing
To test the hypothesis that the locking of blood pressure/heart rate oscillations with respiration is frequency-dependent and not subject-dependent we asked two subjects in a different protocol to breathe at various controlled rates. shows the cross-correlation maps between optical data and physiology during paced breathing in one subject. From left to right the cross-correlation maps of the optical signal with cardiac pulsation, respiration, blood pressure and heart rate are shown. From top to bottom 5 breathing rates are shown, from fast (one breath every 2 sec) to slow breathing (one breath every 10 sec). The y axis reports the time lag between the physiological signal and the optical data (up to ±10sec); the x axis reports the 50 source-detector pairs at 830 nm, the first 25 on the left hemisphere, the second 25 on the right hemisphere, both from front to back, following the detector order shown in . The color bars at the bottom show the amplitude of the cross-correlation. Green represents either zero correlation or discarded channels because of low SNR. From these maps, locking of the blood pressure and heart rate with the respiration for frequencies lower than 0.25 Hz is evident. These measurements also show the spatiotemporal non-uniformity of the cross-correlation between the optical signal and blood pressure/heart rate, as seen during free breathing.
Fig. 8 Cross-correlation maps between optical data and cardiac pulsation, respiration, blood pressure and heart rate during paced breathing in one subject controlling the respiration frequency. In each figure the left represents the left hemisphere, and the (more ...)
3.4. Human subjects –functional measurements
Only 3 of the 8 subjects measured showed statistically significant (p value ≤ 0.05) activation during visual stimulation, 6/10 during cognitive stimulation, and 8/11 during finger tapping. The low success rate of our measurements with visual stimulation was in some cases due to the low SNR for source-detectors in the occipital area. In fact, in 3 subjects with dark or long hair we were not able to achieve a good coupling of the optical probe with the skin. Also in 3 subjects we had to discard several stimulation blocks because of motion artifacts. In these cases the number of blocks remaining was not adequate to obtain statistical significance. We had a relatively good response rate for cognitive stimuli and an even better one for the finger-tapping stimulation protocol. The probe was easier to position over these cortical areas, providing a better SNR, and these stimuli activate more superficial cortical areas than do the visual stimuli used. reports the evoked hemoglobin results for a subject in whom all three of the stimuli showed activation. Oxy-, deoxy-, and total hemoglobin time traces (red, blue, and green, respectively) over a stimulus/rest period are reported for all of the source-detector locations for the three paradigms. The left columns show the block-averaged data where only a bandpass filter between 0.0016 and 0.8 Hz was applied. The right column shows the results with principle components analysis (PCA) analysis where the first spatial eigenvector is eliminated from the optical data. For the finger-tapping stimulation without PCA the activation seems to be everywhere () due to the fact that with this stimulus heart rate and blood pressure increase during the stimulation period, causing a global increase of blood flow. This can be seen in , which reports the cross-correlation of the blood pressure with the stimulus for the three paradigms and shows a good correlation between blood pressure and stimulation periods. After applying the PCA filter to the optical data the activation is localized mostly in the contralateral parietal region (). In the same subject the cognitive stimuli caused a similar increase in blood pressure (), which again masks the evoked hemodynamic response (). After the PCA filter, the activation is localized in the left and right prefrontal regions (). The visual stimuli used do not cause large systemic physiologic response (). In this case, it is not necessary to use PCA on the optical data because the activation area is restricted to the occipital brain regions (). Using the PCA filter would only cause a decrease in the amplitude of the evoked hemoglobin response ().
Fig. 9 Activation maps in a subjects during the three paradigms. Finger tapping, cognitive and visual stimulation are shown top to bottom. The curves represent the block average over a stimulus/rest period (30 sec) of HbO (red), HbR (blue), and HbT (green). (more ...)
Fig. 10 Cross-correlation amplitude of blood pressure and stimulation sequence for the subject shown in during the finger-tapping (a), cognitive (b), and visual (c) paradigms. The blood pressure signal was synchronously measured on a finger of his left (more ...)
The PCA filter helped to better localize the activation in most of our data. It appears to work well for reducing motion artifacts, and for subtracting systemic physiology from the evoked hemoglobin signals as previously discussed by others [1
]. The PCA analysis is effective in these data sets because of the large brain area measured. If we didn’t have the global picture of the hemodynamics over this large brain area, the results in the hand motor cortex alone, for example, would have been difficult to interpret. Problems with the PCA include its tendency to decrease the amplitude of the hemodynamic response in the activated regions, and to propagate noise from noisy channels to all other channels. In addition, the principle components do not necessarily have physiological significance. The result is that their use tends to be ad hoc, working well in some subjects in which the systemic physiological spatiotemporal covariance is well separated from the evoked hemodynamic response, and not so well in others in which the systemic variance is not significantly orthogonal to the evoked hemodynamic response. Finally, preliminary analyses indicate that the physiological spatiotemporal covariance is not necessarily space-time separable. As space-time separability is assumed in the PCA, the enthusiasm for its general implementation is further reduced. Thus, better filters of the systemic physiological fluctuation are desired. Regression with the independent measures of the systemic physiological signals, perhaps combined with Kalman filtering and dynamic state-space modeling is likely to provide improved filtering [18