Estimation of cognitive brain activity in sickle cell disease using functional near-infrared spectroscopy and dynamic systems modeling

• ¹ University of Southern California, Biomedical Engineering, United States
• ² Children's Hospital of Los Angeles, Children’s Center for Cancer, Blood Disease and Bone Marrow Transplantation, United States

Introduction: Frequent vaso-occlusive pain crises in sickle cell disease (SCD) result in silent infarcts as well as permanent damage to critical organs, including the brain. This can cause cognitive impairment and impact patients’ quality of life. The prefrontal cortex (PFC) is known as a control tower for cognitive functions, such as reasoning and memorizing. Functional infrared spectroscopy (fNIRS) is becoming one of popular brain imaging tools for characterizing altered brain responses in prefrontal cortex. However, fNIRS signals also contain non-neuronal fluctuations originating from scalp blood flow and other systemic changes, and this can lead to false conclusions. We propose a systems modeling technique to filter out the confounding influences. Methods and Methods: We hypothesized that the scalp blood flow and the respiration could confound the neurocognitive responses obtained using fNIRS. We designed different levels of n-back memory workload tasks (n [difficulty level] = 0, 1, 2, or 3), to obtain the neurocognitive responses from 14 SCD and 15 controls. During each level of n-back task, the subject was instructed to press a button on the keyboard when the revealed letter was the same as the one shown n-turns ago. The fNIRS was used to record the hemodynamics changes in PFC and scalp, and a nasal CO2 analyzer and a finger photoplethysmogram were used to record CO2 and finger blood flow, respectively. We anticipated an increase in PFC oxygenated hemoglobin content (Hbo) and a decrease in de-oxygenated content (Hbr) while subjects were completing more difficult n-back tasks. After the recording session was over, we used Laguerre-Volterra modeling to estimate the systemic influences originated from scalp blood flow and respiration that had potentially contaminated the fNIRS measurement. We used the end-tidal CO2 and the amplitude of the finger photoplethysmogram signals as two input parameters and the fNIRS Hbo or Hbr signal as an output of the model. We used Laguerre basis functions for estimating impulse responses, which helped minimize the fitting time. We assumed linearity between the two inputs and the fNIRS output. Using the optimal model found here, we estimated the residual signals from all Hbo and Hbr for each channel. The residual signals were saved as ‘filtered’ fNIRS signals assuming they were corrected neuronal activities. The filtered fNIRS signals were grouped by each difficulty of the n-back and averaged. Results and Discussion: There were noticeable influences from CO2 and finger blood flow in the original fNIRS signal (up to ~65% combined), suggesting the signal contamination due to other non-neuronal physiological factors. After we filtered out CO2 and blood flow influences from the original fNIRS, we were able to see the improvements in both the Hbo and Hbr trends, in such that the residual Hbo or Hbr trends became more predictable as the level of difficulty increased (Figure 1). The residual signals showed how our modeling method could help separate neuronal activities from other physiological confounding factors mixed in the fNIRS measurement. Conclusions: We have employed a modeling technique to estimate unknown biological signals that can confound neurocognitive response results obtained from fNIRS. We utilized the end-tidal CO2 and the amplitude of the finger photoplethysmogram signals to remove the confounding influences of arterial PCO2 fluctuations and scalp blood flow from the original fNIRS signal to obtain an improved estimate of PFC activity stimulated by the imposed mental tasks.