ABSTRACT
The identification and quantification of binary blends of refined olive oil with four different extra virgin olive oil (EVOO) varietals (Picual, Cornicabra, Hojiblanca and Arbequina) was carried out with a simple method based on combining visible spectroscopy and non-linear artificial neural networks (ANNs). The data obtained from the spectroscopic analysis was treated and prepared to be used as independent variables for a multilayer perceptron (MLP) model. The model was able to perfectly classify the EVOO varietal (100% identification rate), whereas the error for the quantification of EVOO in the mixtures containing between 0% and 20% of refined olive oil, in terms of the mean prediction error (MPE), was 2.14%. These results turn visible spectroscopy and MLP models into a trustworthy, user-friendly, low-cost technique which can be implemented on-line to characterize olive oil mixtures containing refined olive oil and EVOOs.
Subject(s)
Neural Networks, Computer , Olive Oil/analysis , Food Contamination/analysis , Spectrophotometry, UltravioletABSTRACT
Two of the biggest challenges in medicine today are the need to detect diseases in a noninvasive manner and to differentiate between patients using a single diagnostic tool. The current study targets these two challenges by developing a molecularly modified silicon nanowire field effect transistor (SiNW FET) and showing its use in the detection and classification of many disease breathprints (lung cancer, gastric cancer, asthma, and chronic obstructive pulmonary disease). The fabricated SiNW FETs are characterized and optimized based on a training set that correlate their sensitivity and selectivity toward volatile organic compounds (VOCs) linked with the various disease breathprints. The best sensors obtained in the training set are then examined under real-world clinical conditions, using breath samples from 374 subjects. Analysis of the clinical samples show that the optimized SiNW FETs can detect and discriminate between almost all binary comparisons of the diseases under examination with >80% accuracy. Overall, this approach has the potential to support detection of many diseases in a direct harmless way, which can reassure patients and prevent numerous unpleasant investigations.