In vivo comparison between the principal components analysis and the Karhunen-Loève transform as methods used for the de-noising of laser Doppler reactive hyperemia signals.

This contribution shows the comparison of two methods, the principal components analysis and the Karhunen-Loève transform. Indeed, reactive hyperemia signals obtained with laser Doppler flowmetry are currently used to diagnose peripheral arterial occlusive diseases (PAOD), but they are not noise-free. De-noising of such signals could lead to an improved diagnosis. For this purpose, the principal components analysis and the Karhunen-Loève transform were applied to signals acquired on PAOD and healthy subjects. Our main purpose was to have the two methods undergo a comparison that reveals the capacity of each method to interpret the characteristics of the signals used to make diagnosis. The results show that the use of the Karhunen-Loève transform method is more justified than the principal components analysis whenever we want to reduce the dimensional space of the set of initial data and still preserve the quantitative and relative proportions of the original variances associated to those data representing the laser Doppler flowmetry signal before and after its reconstruction. However, the principal components analysis method is more justified when one or several of the initial data present variances either too insignificant or too important in comparison with the other data.