Assessment of cerebral oxygenation response to hemodialysis using near-infrared spectroscopy (NIRS): Challenges and solutions

Abstract

To date, the clinical use of functional near-infrared spectroscopy (NIRS) to detect cerebral ischemia has been largely limited to surgical settings, where motion artifacts are minimal. In this study, we present novel techniques to address the challenges of using NIRS to monitor ambulatory patients with kidney disease during approximately eight hours of hemodialysis (HD) treatment. People with end-stage kidney disease who require HD are at higher risk for cognitive impairment and dementia than age-matched controls. Recent studies have suggested that HD-related declines in cerebral blood flow might explain some of the adverse outcomes of HD treatment. However, there are currently no established paradigms for monitoring cerebral perfusion in real-time during HD treatment. In this study, we used NIRS to assess cerebral hemodynamic responses among 95 prevalent HD patients during two consecutive HD treatments. We observed substantial signal attenuation in our predominantly Black patient cohort that required probe modifications. We also observed consistent motion artifacts that we addressed by developing a novel NIRS methodology, called the HD cerebral oxygen demand algorithm (HD-CODA), to identify episodes when cerebral oxygen demand might be outpacing supply during HD treatment. We then examined the association between a summary measure of time spent in cerebral deoxygenation, derived using the HD-CODA, and hemodynamic and treatment-related variables. We found that this summary measure was associated with intradialytic mean arterial pressure, heart rate, and volume removal. Future studies should use the HD-CODA to implement studies of real-time NIRS monitoring for incident dialysis patients, over longer time frames, and in other dialysis modalities.

Keywords: Motion artifact removal; cerebral oxygenation; end-stage kidney disease; near-infrared spectroscopy.

Fig. 1.

Diagram of customized headband that integrates a NIRS system for long-term wear.

Fig. 2.

Flow diagram of NIRS signal pre-processing steps.

Fig. 3.

Example of motion artifacts indicated by the magnitude shifts in (top) both wavelengths, which resulted in a decrease in (bottom) both deoxy-hemoglobin (Hb) and oxy-hemoglobin (HbO₂). Changes in the concentrations of HbO₂ and Hb were calculated using the modified Beer-Lambert Law.

Fig. 4.

Example of data processing: (a) raw data, (b) low-pass-filtered data, (c) motion artifacts identified using the MARA algorithm, and (d) comparison between pre- and post-magnitude shift.

Fig. 5.

Figure showing the approach to address motion artifact and variation in between-subjects signal amplitudes using moving SD/mean (top) compared to raw data (bottom).

Fig. 6.

Flow diagram of ischemic condition algorithm (HD-CODA).

Fig. 7.

(a) Segment in black identifies when: β_Hbo2(t) < β_min and ρ(t) < ρ_min. (b) moving correlation between Hb and HbO₂. Based on our chosen parameter values, a data segment with a β < β_min & ρ < ρ_min of at least 2 min duration was defined as an ischemic event (shown in black).

Fig. 8.

Histogram of %drop in (left) session 1, a treatment after a two-day interdialytic interval, and (right) session 2, a treatment after a one-day interdialytic interval. Values represent the proportion of total treatment time spent in cerebral ischemic conditions, as defined in Sec. 2.6.2.

Fig. 9.

Associations between hemodynamic parameters and proportion of treatment time in deoxygenation conditions (i.e., % drop), (left) treatment session 1, and (right) treatment session 2.