Near-Infrared Spectroscopy and Imaging for Investigating Stroke Rehabilitation: Test-Retest Reliability and Review of the Literature

Schematic of neurophysiologic changes expected from a longitudinal rehabilitation and learning study. (A) Idealized response from a region exhibiting linear decreases in activity with over time, assuming a block-design experimental paradigm. Depending on the task or brain region, linear increases or nonlinear (including nonmonotonic) changes may be found. (B) Depiction of an example NIRS response (in arbitrary optical density units) reflecting the idealized response in (A) plus typical physiologic fluctuation observed in NIRS signals (fast oscillations equals cardiac response; slower oscillations equals respiration and/or Mayer waves⁴⁸). Note that the learning-related modulation is smaller than the overall task-related modulation and hence will be harder to detect.

(A) NIRI probe placement and (B) and geometry, where S is the source position (16 total) and D is the detector position (32 total). (C) Example O₂Hb (heavy line) and HHb (light line) responses to 1 run of task performance (shaded regions are finger opposition) for source 6 and detector 12 (left black line in [B], near C3). Notice the substantial task-related modulation. (D) A similar measurement for source 15 and detector 29 (right black line in [B], lateral to C4). Notice the substantial lack of task-related modulation in this region. Abbreviations: Det, (optical) detector; Src, (laser) source; Subj, subject.

The experimental paradigm and illustration of test-retest comparisons. Subjects performed two 5-minute–long runs, each including 8 blocks of finger-opposition task alternated with fixation rest periods. Block-by-block modulation amplitudes were computed by averaging O₂Hb signals (and, separately, HHb signals) for the last 8 seconds of each 16-second task or rest period (short black overbars and underbars, respectively) and subtracting these 2 values (task-rest; thick angled arrows). Test-retest evaluations were performed for 3 lags: adjacent block-pairs (eg, 1 vs 2, 2 vs 3, etc; lag=1), for corresponding blocks across runs (eg, 1 vs 9, 2 vs 10, etc; lag=8), and delayed (block 1 vs 16; lag=15). We also evaluated the effect of averaging the modulation amplitudes across pairs (eg, 1+2 vs 3+4), quadruples (eg, 1+2+3+4 vs 5+6+7+8), and entire runs (17 vs 18).

Histogram of single-block test-retest reliability coefficients (Pearson r values) for adjacent (lag=1) pairs, combined across O₂Hb and HHb.

Effect of lag on Pearson product-moments for single-block test-retest reliability. Mean O₂Hb and HHb Pearson r coefficients are shown for test-retest repetitions of adjacent (lag=1 block; N=252), run (lag=8 blocks; N=144), and delayed (lag=15 blocks, N=18) (see text). Note the estimates for longer lags are based on progressively fewer data points and hence include substantially more variability.

Effect of averaging on Pearson product-moments for NIRS test-retest reliability. Mean O₂Hb and HHb Pearson r coefficients are shown for (A) adjacent comparisons with 1-, 2-, and 4-block averages (along with the mixed-effect model intercept) and (B) run-to-run comparisons, with 1-, 2-, 4-, and 8-block averages (along with the intercept).