Artificial neural networks (ANN) was evaluated and compared with Response Surface Model (RSM) results using growth response data for E.coli O157:H7 as affected by 5 variables: pH, sodium chloride, and nitrite concentrations, temperature, and aerobic/anaerobic conditions. The best ANN obtained, where the 2 kinetic parameters, growth rate and lag-time, were estimated jointly, contained 17 parameters and displayed a slightly lower Standard Error of Prediction (% SEP) than those obtained with RSM. Mathematical lag-time validation with additional data gave a lower %SEP for ANN (18%) than for RSM (27%), although growth-rate values were the same (22%). ANN thus should provide the innovative possibility of obtaining a single predictive model for the estimation of several kinetic parameters.

m e shelf-life of vacuum packed, sliced, cooked chicken-breast based on sensory and microbial changes as afunction of temperature (2.3, 6.5, 10, 13.5 and I7.7C) was determined. Sensory evaluation and a microbiological study charted the development of lactic acid and psychotropic bacteria and of Brochothrix thermosphacta. Six different sensory methods were used to estimate product shelf-life; of these, the method based on average smell and taste was deemed the most suitable, since these parameters had a greater impact on shelf-life. From a microbiological point of view, mean shelf-life times were estimated at each temperature and compared with the estimates of the tasting panel. In the samples stored at the three lowest study temperatures (2.3, 6.5 and IOC), lactic acid and psychotropic bacteria counts of ld-l@ cfu/g were not achieved; this agrees with the absence of sensory rejection at the end of the experiment. At 13.5 and I7.7C, mean shelf-life estimated microbiologically was shorter than that estimated using sensory methods. This difference, which here amounted to at least 8 days, is due to the so-called "delayed change", reported in previous experiments with cooked meat products.

No abstract