Quantitative comparison of correction techniques for removing systemic physiological signal in functional near-infrared spectroscopy studies

Figures

• 

  Fig. 1. FNIRS optodes configuration. Blue solid-line and green dotted-line represent the LD (n=22) and SS (n=8) channels, respectively. Detector 4 coincides with the Cz location in the International 10–20 System.

• 

  Fig. 2. In this work, many combinations of preprocessing and statistical models (GLM) were applied to the three types of simulated data. (a) The 60 (3×4×5) combinations of processing that were used in this study. Examples of (b) the experimental resting and (c) BH data are shown, which were used to generate the three types of semisynthetic data sets by adding known evoked responses to a subset of the experimental data. (c) The timing of the stimulus events used in the BH-locked and BH-random simulations.

• 

  Fig. 3. Comparison of (a)–(c) the AUC from (d)–(f) sensitivity–specificity ROC curve. In every panel, six selected different processing pipelines are investigated. (a), (d) Resting data; (b), (e) BH with random stimulus marking; and (c), (f) BH with locked stimulus marking.

• 

  Fig. 4. Comparison of OLS and AR-IRLS of control type-I error reports from three different data sets using selected six different processing pipelines. (a) Resting data, (b) BH with random stimulus marking, and (c) BH with locked stimulus marking.

• 

  Fig. 5. Comparison of AUC from ROC analysis at various CNRs of 0.5, 0.7, 1.0, and 2.0 from selected four different processing pipelines. (a) Resting data, (b) BH with random stimulus marking, and (c) BH with locked stimulus marking.

• 

  Fig. 6. Comparison of sensitivity–specificity of AUC without preprocessing using various nearest SS channels for six selected processing. All these processing have been implemented using three datasets: (a) resting, (b) BH (random), and (c) BH (locked) data.

• 

  Fig. 7. Comparison of ROC performances (AUC values) with respect to SCI values. The AUC for the AR-IRLS + SS model using SS data of varied quality. Each bar shows the AUC for data using SS channels with a specific SCI (bins of ±0.05).

• 

  Fig. 8. Comparison of AUC from ROC analysis at various cutoff frequencies of LPF ([0, 0.5] Hz) and HPF ([0, 0.005, 0.008, 0.016, 0.032, 0.064] Hz) from (a), (b) HbO2 and (c), (d) Hb. In every panel, BH (random) data using AR-IRLS with/without SS-regressor at various levels of HPF are investigated.