Physiological measures of cognitive load have become more accessible with advancements in neuro-imaging devices and wearable technology, and provide quantifiable data that previous measurement tools (for example, surveys) could not. As such, finding adequate new physiological measures of cognitive load is important [1]. Zhang et al. present a real-time quantitative measurement method for cognitive load via blood volume changes (photoplethysmogram, or PPG). Existing measurement methods are more invasive and time-insensitive, for example, an electrocardiogram (ECG) only requires a finger sensor that can take a reading of every pulse and is able to identify low and high load tasks. They validate their method by having 16 participants perform a series of n-back memory tests. Half of these tests were used to train many machine learning algorithms to classify the cognitive load of each session, while the other half were used to test the algorithm’s classification accuracy. Successful classification indicates that PPG would be a good method for measuring cognitive load.

The novelty of real-time single waveform measurements appears useful when there are large differences in cognitive load (zero versus two-back tests), but only slightly better than random for lower loads (zero versus one-back), even when these are statistically different based on root-mean-square error (RMSE) and fraction correct measures. Applications of cognitive load measurements focus on executive function tasks that mimic real-world tasks; only being able to reliably differentiate very different cognitive loads significantly reduces its practical applications. The classification results are more promising with the two-minute dataset; however, given that on a human timescale two minutes is not significantly different than ECGs (three to five minutes) and no data is offered to compare their relative reliability, it is not clear that PPG is an improvement. Clearly more research is needed.

It is important to note that the experimental setup is very similar to previous work [2], which is great for replication purposes. Unfortunately, the authors of neither paper make their data publicly accessible. The statistical analysis of [2] is also more sophisticated than the current paper. The main improvement, in practice, is the reduced number of sensors needed as well as their lower invasiveness. As cognitive load measurement has applications in task development, performance, ergonomics, and training, a discussion of wider applicability was sorely missed. The paper contains a disturbing number of typographical errors. The line of work is interesting, but readers are better off reading [2].