Threshold optimization in separating cortical and extracerebral hemodynamics using principal component analysis

When trying to differentiate between hemodynamic cortical and extracerebral signals identified by devices used to detect cortical activity, statistical methods such as principal component analysis (PCA) are commonly employed as alternative approaches to using short separation measurements to reduce the influence of extracerebral hemodynamics. PCA requires a threshold value to separate cortical and extracerebral signals; however, existing methods often rely on fixed thresholds that fail to account for inter-individual variability and differences in experimental design, potentially leading to over- or under-correction. Rather than introducing a novel extracerebral hemodynamics removal method, the present study aims to optimize the use of existing methodologies. Specifically, we proposed a method to optimize the threshold that differentiates cortical from extracerebral hemodynamics in PCA-based analyses.

Each of the four analyses were applied to a dataset obtained from older participants performing a verbal n-back task: (1) no correction (NC), (2) short separation regression (SSR), (3) PCA with our proposed threshold optimization (PCA_opt), and (4) PCA with the individual maximum as threshold (PCA_max). Bayesian t-tests were then conducted to evaluate the equivalence between SSR and PCA_opt.

NC displayed the strongest cortical activation, PCA_max the weakest. SSR and PCA_opt produced intermediate results, and Bayesian t-tests revealed that the BF₀₁ values for most of the channels were greater than 3.0, whereas no channels exhibited corresponding BF₁₀ values exceeding 3.0.

Optimizing the threshold for separating cortical and extracerebral hemodynamics is a practical and effective strategy when using PCA as an alternative to short-separation measurements. This approach enables appropriate correction even in the absence of short-separation channels.