PubMed2000
The problem of making quantitative assessments of the risks associated with human exposure to toxic contaminants in the environment is a pressing one. This study demonstrates the capability of a new computational technique involving the use of fuzzy logic and neural networks to produce realistic risk assessments. The systematic analysis of human exposure involves the use of measurements and models, the results of which are sometimes used in regulatory decisions or in the drafting of legislation. Because of limited scientific understanding, however, interpretation of models often involves substantial uncertainty. Extensive measurement programs can be very expensive. The high complexity and inherent heterogeneity of exposure analysis is still a major challenge. The approach to this challenge tested here is to use a new model incorporating sophisticated artificial intelligence algorithms. Exposure assessment often requires that a number of factors be evaluated, including exposure concentrations, intake rates, exposure times, and frequencies. These factors are incorporated into a system that can "learn" the relevant relationships based on a known data set. The results can then be applied to new data sets and thus be applied widely without the need for extensive measurements. In this analysis, an example is developed for human health risk through inhalation exposure to benzene from vehicular emissions in the cities of Auckland and Christchurch, New Zealand. Risk factors considered were inhaled contaminant concentration, age, body weight, and activity patterns of humans. Three major variables affecting the inhaled contaminant concentration were emissions (mainly from motor vehicles), meteorology (wind speed, temperature, and atmospheric stability), and site factors (hilly, flat, etc.). The results are preliminary and used principally to demonstrate the technique, but they are very encouraging.