|SEO, YOUNGWOOK - US Department Of Agriculture (USDA)|
Submitted to: Transactions of the ASABE
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/12/2017
Publication Date: 1/15/2018
Citation: Seo, Y., Park, B., Yoon, S.C., Lawrence, K.C., Gamble, G.R. 2018. Morphological image analysis for foodborne bacteria classification. Transactions of the ASABE. 61(1): 5-13.
Interpretive Summary: Among serious foodborne outbreaks during past decades, Salmonella had the most infections and incidence cases accounting for over 15% of foodborne outbreaks. Specifically, each year more than million people are sickened by Salmonella in the United States with approximately 200,000 cases from poultry alone, resulting in a significant cost of foodborne illness. Therefore there is a need to reduce foodborne illnesses, especially in poultry products. Although current traditional culturing and detecting methods are still the golden standard for foodborne pathogen identification, these methods are time consuming and labor intensive. Advanced optical methods such as hyperspectral microscope imaging (HMI) has the potential to evaluate foodborne pathogens for enhancing the presumptive-positive screening method, by reducing labor and increasing speed of sample analysis. In this study, we developed HMI methods to identify foodborne pathogenic bacteria with minimum sample preparation using morphological characteristics of bacterial cells.
Technical Abstract: Previous hyperspectral imaging methods for food safety applications focused on spectral data analysis to understand spectral characteristics relevant to safety of food and agricultural commodities. However, the exploration with spatial information including morphological characteristics such as size, orientation, shape, color, and texture has not been actively investigated for hyperspectral imaging. In this study, image processing techniques were applied for segmentation and extraction of morphological features from five different foodborne bacteria including Salmonella, E. coli, Enteobacter, Listeria, as well as Enterococcus. Using selected morphological features including mean, standard deviations, maximum axis length, minimum axis length, orientation, equivalent diameter, solidity, extent, perimeter, eccentricity, equivalent circular diameter, and axis ratio extracted by auto-cell segmentation algorithm, bacteria were differentiated and classified with generalized linear model and multivariate analysis methods with algorithms of decision tree and support vector machine. Overall classification accuracies to identify gram-negative from gram-positive bacteria were 84.5%. However, the classification accuracy to identify five species was only 75.1%. Thus, a hyperspectral image analysis with morphological features in the spatial domain is limited to identify bacterial species, so that additional spectral feature analysis is needed to improve classification accuracy.