Objective selection of signals for assessment of cerebral blood flow autoregulation in neonates

A number of different system identification techniques have been proposed to assess dynamic cerebral autoregulation in critically ill patients. From these methods, the response to a standard stepwise change in blood pressure can be estimated. Responses lacking physiological consistency are a common occurrence and could be the consequence of particular system identification procedures or, alternatively, caused by measurements with a poor signal-to-noise ratio. A multi-observer approach was adopted in this paper to classify cerebral blood flow velocity (CBFV) step responses to spontaneous changes in arterial blood pressure in a group of 43 neonates with a mean gestational age of 33.7 weeks (range 24-42 weeks) and a mean birthweight of 1,980 g (range 570-3,910 g). Three experienced observers independently analysed the estimated step responses in 191 recordings each lasting 100 s; for an autoregressive (ARX) model, 124 (65%) of the step responses were accepted by at least two of the three observers. Two other system identification methods, transfer function analysis and the moving average Wiener-Laguerre model, gave 90 (45%) and 98 (51%) acceptable responses, respectively. Only 54 epochs (28%) were accepted with all three methods. With 88 (46%) responses rejected by at least two methods, it can be concluded that signal quality was the main reason for nonphysiological step responses. To avoid the need for subjective visual selection, an automatic procedure for classifying step responses was implemented leading to sensitivities and specificities in the range 85-90%, with respect to the agreement with subjective evaluations. Objective selection of CBFV step responses is thus feasible and could also be adapted for other physiological measurement techniques relying on system identification methods.