A rapid and non-destructive detection of Escherichia coli on the surface of fresh-cut potato slices and application using hyperspectral imaging

Highlights

• E. coli on the fresh-cut potato slices was successfully investigated by HSI.

• BP-NN model showed higher accuracy and robustness for E. coli detection.

• The feasibility of HSI for the practice application was confirmed for the first time.

Abstract

The contamination of foodborne Escherichia coli in fresh-cut products has become a major problem of public health around the world, so that early and rapid detection of contamination is crucial. This study explored the potential of hyperspectral imaging (HSI) measurement of contamination on the surface of fresh-cut potato slices in visible-near infrared (Vis-NIR, 400-1000 nm) region. Four preprocessing methods and the genetic algorithm (GA) were explored to handle spectral data and select characteristic wavelengths so as to establish linear and non-linear regression models. The performance of the back-propagation neural network (BP-NN) model based on full-spectrum was satisfactory, with an overall accuracy of 97.6 % and residual predictive deviation (RPD) of 6.7. Based on the BP-NN model, the research successfully explored the optimum treatment time (20 min) of a non-thermal and environmental-friendly method to inactivate the E. coli on the surface of fresh-cut potato slices, thus confirming the potential application of HSI for the first time. The overall results showed that HSI could provide a rapid and non-destructive approach for the detection of foodborne pathogens on the surface of fresh-cut products.

Introduction

With the acceleration of people’s pace of life and pursuit of convenience and health, the demand for fresh-cut fruit and vegetable is increasing dramatically (Liu et al., 2019), and a large number of processed potatoes have appeared in markets for more than a decade (Zhou et al., 2019). However, some foodborne pathogens are faster to grow on the exposed surface of fresh-cut products caused by mechanical injury after minimal processing, including Escherichia coli (Yuan et al., 2019), Salmonella (Cuggino et al., 2020), Listeria monocytogenes (Scollard et al., 2016), and Penicillium expansum (Rios De Souza et al., 2020), which would pose a threat to human health or exacerbate the decay of fresh-cut products to reduce their edible value and shorten the shelf life, so early detection for pathogens is necessary (Condurso et al., 2020). Traditional biological culture-dependent methods often need long-time and laborious work, involving complex processes of sample preparation, serial dilution, plating on a suitable medium and a certain incubation time to obtain visible colonies (Hameed et al., 2018). Recent researchers have focused on developing rapid and accurate techniques to detect pathogens in food, such as polymerase chain reaction (PCR), biosensor and enzyme-linked immunosorbent assay (ELISA), but these methods are often technically complicated, requiring well-trained specialists and inevitably damaging samples (Kang, 2019; Ye et al., 2019; Nagaraj et al., 2016).

Hyperspectral imaging (HSI) can collect data of three-dimensional patterns, two spatial and one wavelength dimensions to obtain related qualitative and quantitative information as a rapid and non-destructive detection method (Bonah et al., 2019; He and Sun, 2015). Compared with complex and destructive traditional detection approaches, the advantage of HSI becomes outstanding because this method requires little or even no sample preparation and can measure several chemical compositions and quality attributes simultaneously (Zhao et al., 2020). Recent studies have investigated HSI as a non-destructive and effective method for prediction of fruit quality attributes such as ripeness (Munera et al., 2017), pH (Li et al., 2018a,2018b), soluble solids content (SSC) (Zhang et al., 2019), firmness (Li et al., 2018a,2018b), chilling injury (Babellahi et al., 2020), and fungal infections (Siedliska et al., 2018).

HSI has also been applied to evaluate pathogens on food, based on the assumption that the metabolic activity of pathogens would lead to biochemical changes and metabolic byproducts, which could provide characteristic fingerprints that potentially indicated the contamination in food (Wang et al., 2018). For example, Cheng and Sun (2015) developed a multiple linear regression (MLR) model with residual predictive deviation (RPD) and the coefficient of determination of prediction (R²_P) was 5.22 and 0.870 by the HSI method in the 400-1000 nm spectral range to measure E. coli in grass carp. Besides, Huang et al. (2013) exploited a non-destructive detection of total viable count (TVC) in pork with a back-propagation artificial neural network (BP-ANN) algorithm which built a prediction model (R²_P = 0.8308, the root mean square error of prediction (RMSEP) = 0.243) by HSI (400-1000 nm). Although the above researchers have confirmed the feasibility of the HSI system in several meat products, only few studies have reported the experimental result in fresh-cut fruit and vegetables. For example, Siripatrawan et al. (2011) assessed E. coli contamination in packaged fresh spinach by HSI ranging from 400-1000 nm combined with PCA and ANN to predict the number of E. coli with R² of 0.97.

Our previous research has shown that photodynamic treatment (PDT) has a satisfactory performance to inactivate E. coli on fresh-cut apples and is considered as a very promising approach for antimicrobial approach (Tao et al., 2019). In this study, using HSI to establish a non-destructive method and explore the optimal illumination time for PDT was the first step to explore the feasibility of practical application based on the HSI system. The specific research objectives were: (1) to use 400-1000 nm hyperspectral imaging method to image the fresh-cut potato slices with E. coli on the surface; (2) to build the best model based on full-spectrum and characteristic wavelength by genetic algorithm (GA); (3) to obtain optimum illumination time of PDT on fresh-cut potato slices with best inactivation effect using the established optimal model.