Evaluating artificial neural networks for predicting minute ventilation and lung deposited dose in commuting cyclists

Evidence linking personal air pollution exposure to adverse human health impacts is well reported in literature. Commuting in urban traffic micro-environments often leads to a large proportion of total daily exposure and uptake. Cyclists in particular, due to their elevated physical exertion levels and ventilatory parameters, experience higher uptakes of air pollution while commuting relative to less active commuters. A model for predicting minute ventilation of cyclist commuters in the field was developed, and PM10 lung deposited doses were predicted based on this. Sixty healthy volunteers were recruited. Minute ventilation, heart rate, personal air pollution exposure, local meteorological conditions, GPS acquired cycling speed and road topography were continuously monitored during sampling protocols. An artificial neural network (ANN) model for predicting minute ventilation was developed based on these variables and subject characteristics. Predicted values were regressed against measured minute ventilation. A Generalised Additive Model (GAM), a Partial Least Squares (PLS) model and three empirical minute ventilation models were tested in the same manner. The ANN, GAM and PLS predicted minute ventilation levels showed better agreement with measured minute ventilation values (R2=0.82, 0.74 and 0.56 respectively) than the empirical models (R2 values ranging from 0.36 to 0.43). The average percentage error of the ANN modelled minute ventilation (2.5±20.2%) was smallest of all models tested. Lung deposited doses of air pollution were calculated using a human respiratory tract model. Doses calculated utilising ANN modelled minute ventilation demonstrated the best agreement with the lung deposited dose determined using measured minute ventilation. Cycling scenarios were investigated in terms of ANN modelled minute ventilation levels and PM10 lung deposited doses. This study presents a novel method of indirectly measuring cyclists׳ breathing rates in an urban outdoor setting and will have applications in assessing accurate air pollution uptake in cyclists in future research.