A Model-Based Artificial Neural Network to Predict Blood Acidity in Patients under Cardiopulmonary Bypass

Abstract

Blood acidity (pH) is an important parameter to evaluate the acid-base status in patients during cardiopulmonary bypass (CPB). To date, pH measurement is done using blood gas analysis through intermitted blood sampling or continuously using an expensive disposable sensor. In this paper, we explore the capability of measuring arterial pH from blood gases to reduce the cost of the monitoring system. The study includes 353 blood gas analyses collected during CPB. Blood gas analyses from patients with renal failure were excluded from the study. The study includes designing a two-layered-feed-forward, backpropagation neural network to measure pH from the arterial partial pressure of carbon dioxide (P <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">a</inf> CO <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> ). Further, regression methods were also explored for comparison and parameter selection. The network model accuracy was evaluated using the root mean square error (RMSE), coefficient of determination (R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> ), Akaike’s criteria, and post-regression analysis. The results showed that the neural network model has higher accuracy in predicting pH with a RMSE of about 0.031 and R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 81.5%. The present findings highlighted that arterial pH could be estimated from P <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">a</inf> CO <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> . By applying it clinically using oxygenator exhaust capnometry, perfusionists can manage the patient respiratory acidity earlier.