Detection of aflatoxin B1 on corn kernel surfaces using visible-near infrared spectroscopy

Authors: TAO, FEIFEI - Mississippi State University; YAO, HAIBO - Mississippi State University; HRUSKA, ZUZANA - Mississippi State University; Liu, Yongliang; Rajasekaran, Kanniah - Rajah; Bhatnagar, Deepak

Submitted to: Journal of Near Infrared Spectroscopy
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/23/2019
Publication Date: 12/18/2019
Citation: Tao, F., Yao, H., Hruska, Z., Liu, Y., Rajasekaran, K., Bhatnagar, D. 2019. Detection of aflatoxin B1 on corn kernel surfaces using visible-near infrared spectroscopy. Journal of Near Infrared Spectroscopy. 28(2):59-69. https://doi.org/10.1177/0967033519895686.
DOI: https://doi.org/10.1177/0967033519895686

Interpretive Summary: The differences existing between the endosperm and germ sides of corn kernels may have negative effect on spectroscopic detection of AFB1-contaminated corn kernels, with both sides of corn kernels as background food matrices. Previous studies have shown the possibility of detecting surface aflatoxin B1 (AFB1)-contaminated corn kernels from the germ side, but it is still not clear whether it is possible to establish a common model based on spectroscopic technology, when both sides of corn kernels are included. Therefore, the artificially-contaminated corn kernels were prepared on both sides of corn kernels with 7 contamination levels, namely, 0 (control), 10, 20, 50, 100, 500 and 1000 ppb. Different spectral preprocessing methods including standard normal variate (SNV), first derivative and second derivative transformations were applied on the original absorbance spectra, and the 2-class, 3-class and 7-class types of partial least squares discriminant analysis (PLS-DA) classification models were established. With both classification thresholds of 20 ppb and 100 ppb, the 2-class PLS-DA models achieved the best overall accuracies of 100.0%. The best 3-class classification model obtained the overall accuracy of 95.7% using the preprocessed spectra over 1120-2470 nm. The 7-class classification models did not perform as well. However, further statistical analyses yielded over 70.0% overall accuracies. Comparison of all the classification modeling results using different combinations of spectral range and spectral preprocessing, indicates that the spectral range 1120-2470 nm and derivative transformations (first and second derivatives) performed better at identifying corn kernels surface-contaminated with AFB1.

Technical Abstract: In this study, we utilized visible/near-infrared (Vis/NIR) spectroscopy over the spectral range of 400-2500 nm to detect surface contamination of corn kernels with aflatoxin B1 (AFB1). The artificially-contaminated samples were prepared by dropping known amounts of AFB1 standard in methanol, onto corn kernel surface to achieve different contamination levels of 10, 20, 50, 100, 500 and 1000 ppb. Both endosperm and germ sides of corn kernels were used for artificial contamination, and a total of 210 contaminated and control kernels were scanned with the Vis/NIR spectroscopy in reflectance mode. The 2-class and 3-class classification models achieved accurate overall accuracies in classifying the surface AFB1-contaminated and healthy corn kernels. With both classification thresholds of 20 ppb and 100 ppb, the 2-class PLS-DA models achieved the best overall accuracies of 100.0%. The best 3-class classification model obtained the overall accuracy of 95.7% using the preprocessed spectra over range II (1120-2470 nm). The 7-class classification models did not perform as well. Only the spectra preprocessed by the second derivative transformation and 25-point S-G smoothing attained over 70.0% overall accuracies. Comparison of all the classification modeling results using different combinations of spectral range and spectral preprocessing, indicates that the spectral range II and derivative transformations (first and second derivatives) performed better at identifying corn kernels surface-contaminated with AFB1.