Classification of aflatoxin contaminated single corn kernels by ultraviolet to near infrared spectroscopy

Xianbin Cheng, Andrea Vella, Matthew Jon Stasiewicz

Research output: Contribution to journalArticle

Abstract

Aflatoxin contamination in corn poses threats to consumer food safety and grower economic stability. Current industrial methods for aflatoxin management in corn focus on the bulk aflatoxin level, which can lead to either acceptance of lots with contaminated corn kernels (consumer food safety risk) or rejection of lots with mostly harmless corn kernels (grower economic loss). This dilemma may be resolved by utilizing spectroscopy to classify single corn kernels. Hence, our research aims to investigate the potential of using a custom-built UltraViolet-Visible-Near InfraRed spectroscopy system (UV–Vis–NIR) to classify single corn kernels by aflatoxin level. Single kernels from cobs inoculated with aflatoxin-producing Aspergillus flavus (240 kernels) and uninoculated cobs (240 kernels) were i) scanned individually for reflectance from 304 nm to 1086 nm by an increment of 0.5 nm; ii) ground; iii) measured for aflatoxin by ELISA. Using the spectra and the aflatoxin concentration, a random forest model was trained on 80% of the kernels to classify single corn kernels above or below 20 ppb of aflatoxin and was tested on the remaining 20% of the kernels. Among 480 kernels, 374 kernels had <20 ppb of aflatoxin and 106 kernels had ≥20 ppb of aflatoxin. The random forest model had a sensitivity of 87.1% and specificity of 97.7% in the training set and a sensitivity of 85.7% and specificity of 97.3% in the test set, which is higher than previous models where kernels were in motion and comparable to models where kernels were stationary. Spectral regions around 390, 540, and 1050 nm are found to be important for classification. This study demonstrated the custom-built UV–Vis–NIR spectroscopy system showed considerable potential in classifying single corn kernels by aflatoxin level while the kernels are in motion.

Original languageEnglish (US)
Pages (from-to)253-261
Number of pages9
JournalFood Control
Volume98
DOIs
StatePublished - Apr 1 2019

Fingerprint

Aflatoxins
Near-Infrared Spectroscopy
near-infrared spectroscopy
aflatoxins
Zea mays
corn
seeds
Food Safety
Spectrum Analysis
ultraviolet-visible spectroscopy
Economics
Sensitivity and Specificity
Aspergillus flavus
food safety
spectroscopy
growers
economics
Enzyme-Linked Immunosorbent Assay

Keywords

  • Aflatoxin
  • Classification
  • Corn
  • ELISA
  • Random forest
  • Spectroscopy

ASJC Scopus subject areas

  • Biotechnology
  • Food Science

Cite this

Classification of aflatoxin contaminated single corn kernels by ultraviolet to near infrared spectroscopy. / Cheng, Xianbin; Vella, Andrea; Stasiewicz, Matthew Jon.

In: Food Control, Vol. 98, 01.04.2019, p. 253-261.

Research output: Contribution to journalArticle

@article{6b6ee663baab4d6ca96aed9807acc553,
title = "Classification of aflatoxin contaminated single corn kernels by ultraviolet to near infrared spectroscopy",
abstract = "Aflatoxin contamination in corn poses threats to consumer food safety and grower economic stability. Current industrial methods for aflatoxin management in corn focus on the bulk aflatoxin level, which can lead to either acceptance of lots with contaminated corn kernels (consumer food safety risk) or rejection of lots with mostly harmless corn kernels (grower economic loss). This dilemma may be resolved by utilizing spectroscopy to classify single corn kernels. Hence, our research aims to investigate the potential of using a custom-built UltraViolet-Visible-Near InfraRed spectroscopy system (UV–Vis–NIR) to classify single corn kernels by aflatoxin level. Single kernels from cobs inoculated with aflatoxin-producing Aspergillus flavus (240 kernels) and uninoculated cobs (240 kernels) were i) scanned individually for reflectance from 304 nm to 1086 nm by an increment of 0.5 nm; ii) ground; iii) measured for aflatoxin by ELISA. Using the spectra and the aflatoxin concentration, a random forest model was trained on 80{\%} of the kernels to classify single corn kernels above or below 20 ppb of aflatoxin and was tested on the remaining 20{\%} of the kernels. Among 480 kernels, 374 kernels had <20 ppb of aflatoxin and 106 kernels had ≥20 ppb of aflatoxin. The random forest model had a sensitivity of 87.1{\%} and specificity of 97.7{\%} in the training set and a sensitivity of 85.7{\%} and specificity of 97.3{\%} in the test set, which is higher than previous models where kernels were in motion and comparable to models where kernels were stationary. Spectral regions around 390, 540, and 1050 nm are found to be important for classification. This study demonstrated the custom-built UV–Vis–NIR spectroscopy system showed considerable potential in classifying single corn kernels by aflatoxin level while the kernels are in motion.",
keywords = "Aflatoxin, Classification, Corn, ELISA, Random forest, Spectroscopy",
author = "Xianbin Cheng and Andrea Vella and Stasiewicz, {Matthew Jon}",
year = "2019",
month = "4",
day = "1",
doi = "10.1016/j.foodcont.2018.11.037",
language = "English (US)",
volume = "98",
pages = "253--261",
journal = "Food Control",
issn = "0956-7135",
publisher = "Elsevier BV",

}

TY - JOUR

T1 - Classification of aflatoxin contaminated single corn kernels by ultraviolet to near infrared spectroscopy

AU - Cheng, Xianbin

AU - Vella, Andrea

AU - Stasiewicz, Matthew Jon

PY - 2019/4/1

Y1 - 2019/4/1

N2 - Aflatoxin contamination in corn poses threats to consumer food safety and grower economic stability. Current industrial methods for aflatoxin management in corn focus on the bulk aflatoxin level, which can lead to either acceptance of lots with contaminated corn kernels (consumer food safety risk) or rejection of lots with mostly harmless corn kernels (grower economic loss). This dilemma may be resolved by utilizing spectroscopy to classify single corn kernels. Hence, our research aims to investigate the potential of using a custom-built UltraViolet-Visible-Near InfraRed spectroscopy system (UV–Vis–NIR) to classify single corn kernels by aflatoxin level. Single kernels from cobs inoculated with aflatoxin-producing Aspergillus flavus (240 kernels) and uninoculated cobs (240 kernels) were i) scanned individually for reflectance from 304 nm to 1086 nm by an increment of 0.5 nm; ii) ground; iii) measured for aflatoxin by ELISA. Using the spectra and the aflatoxin concentration, a random forest model was trained on 80% of the kernels to classify single corn kernels above or below 20 ppb of aflatoxin and was tested on the remaining 20% of the kernels. Among 480 kernels, 374 kernels had <20 ppb of aflatoxin and 106 kernels had ≥20 ppb of aflatoxin. The random forest model had a sensitivity of 87.1% and specificity of 97.7% in the training set and a sensitivity of 85.7% and specificity of 97.3% in the test set, which is higher than previous models where kernels were in motion and comparable to models where kernels were stationary. Spectral regions around 390, 540, and 1050 nm are found to be important for classification. This study demonstrated the custom-built UV–Vis–NIR spectroscopy system showed considerable potential in classifying single corn kernels by aflatoxin level while the kernels are in motion.

AB - Aflatoxin contamination in corn poses threats to consumer food safety and grower economic stability. Current industrial methods for aflatoxin management in corn focus on the bulk aflatoxin level, which can lead to either acceptance of lots with contaminated corn kernels (consumer food safety risk) or rejection of lots with mostly harmless corn kernels (grower economic loss). This dilemma may be resolved by utilizing spectroscopy to classify single corn kernels. Hence, our research aims to investigate the potential of using a custom-built UltraViolet-Visible-Near InfraRed spectroscopy system (UV–Vis–NIR) to classify single corn kernels by aflatoxin level. Single kernels from cobs inoculated with aflatoxin-producing Aspergillus flavus (240 kernels) and uninoculated cobs (240 kernels) were i) scanned individually for reflectance from 304 nm to 1086 nm by an increment of 0.5 nm; ii) ground; iii) measured for aflatoxin by ELISA. Using the spectra and the aflatoxin concentration, a random forest model was trained on 80% of the kernels to classify single corn kernels above or below 20 ppb of aflatoxin and was tested on the remaining 20% of the kernels. Among 480 kernels, 374 kernels had <20 ppb of aflatoxin and 106 kernels had ≥20 ppb of aflatoxin. The random forest model had a sensitivity of 87.1% and specificity of 97.7% in the training set and a sensitivity of 85.7% and specificity of 97.3% in the test set, which is higher than previous models where kernels were in motion and comparable to models where kernels were stationary. Spectral regions around 390, 540, and 1050 nm are found to be important for classification. This study demonstrated the custom-built UV–Vis–NIR spectroscopy system showed considerable potential in classifying single corn kernels by aflatoxin level while the kernels are in motion.

KW - Aflatoxin

KW - Classification

KW - Corn

KW - ELISA

KW - Random forest

KW - Spectroscopy

UR - http://www.scopus.com/inward/record.url?scp=85057070421&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85057070421&partnerID=8YFLogxK

U2 - 10.1016/j.foodcont.2018.11.037

DO - 10.1016/j.foodcont.2018.11.037

M3 - Article

VL - 98

SP - 253

EP - 261

JO - Food Control

JF - Food Control

SN - 0956-7135

ER -