Quantifying autoregulation from estimated model parameters: an optimization approach (1184.8)

In order to quantify cerebral autoregulation (CA), one of the most commonly used techniques is to fit a linear model to the recorded signals, with arterial blood pressure (ABP) as input and cerebral blood flow velocity (CBFV) as output. Then parameters are extracted from the model to quantify CA: for example average phase or gain in a narrow frequency band, or features of the predicted step‐response. We propose an optimized method to distinguish between CA during normocapnia and hypercapnia. In 27 adults (18‐55 years old) at rest, CBFV and ABP were recorded simultaneously during normocapnia and then hypercapnia induced by breathing 5% CO2 in air. Recordings were repeated on a separate day, giving a total of 108 recordings of approximately 5 minutes each, all resampled at 1 Hz. The interrelationship between ABP and CBFV was modelled using a linear finite‐impulse‐response filter (7 coefficients). Multiple linear regression of these coefficients was used to predict normocapnia (1) or hypercapnia (0), achieving an area under the ROC curve of 0.91, considerably better than ARI (0.81) and slightly higher than for phase‐lead at 0.1 Hz (0.88). The method provides a new, data‐driven and optimized parameter for assessing autoregulation. This work is being extended to include non‐linear classifiers (support vector machines). Grant Funding Source : Supported by EPSRC (UK) and Central University, Chile