Table 1.
Summary of research studies that report single corn kernel aflatoxin and fumonisin contamination detection using different analytical methods.
| Analytical Method used | Mycotoxin tested | Contaminated corn source | Kernel Motion State | Measurement Type | Spectral Analysis method* | Classification Accuracies and Major Results | Reference |
|---|---|---|---|---|---|---|---|
| Liquid Chromatography | Total Fumonisin | Contaminated and uncontaminated samples obtained from farmers | - | Tandem mass spectrometry | - | 39% of kernels (155/400) were contaminated with 1.84–1428 mg/kg fumonisin. Only 4% were above 100 ppb fumonisin and removal of these kernel reduced average fumonisin content by 71%. | [22] |
| Fluorescence Imaging | Fumonisin FB1 |
In field inoculated samples | - | Fluorescence Polarization | - | Fumonisin concentration was correlated with fluorescence (r2 = 0.85–0.88). | [29] |
| Fluorescence Imaging/Liquid Chromatography | Fumonisin FB1, FB2 |
Obtained from commercial sources | - | Fluorescence detection (FD), mass spectrometry | - | Data validation method reproducibility ≤ 15.9% and recovery 78–110%. | [24] |
| Fluorescence Imaging | Total Aflatoxin | In field inoculated samples | Stationary | Fluorescence emission | Linear regression | 84% classification accuracy at threshold of 20ppb, and 86% at classification threshold of 100ppb. | [26] |
| Fluorescence/Reflectance Imaging | Total Aflatoxin | In field Inoculated samples | Stationary | Fluorescence, reflectance visible near-infrared (VNIR) | KNN | 84% sensitivity and 96% specificity for classification model at a threshold of 20 ppb. | [42] |
| Fluorescence/Reflectance Imaging | Total Aflatoxin | In Field inoculated samples | Stationary | Fluorescence, reflectance visible near-infrared (VNIR) | PCA, LS-SVM, KNN | Threshold values of 20 and 100 ppb were used. Classification models: 92% sensitivity and 96% at threshold of 100 ppb; 89% sensitivity and 96% specificity threshold of 20 ppb. | [28] |
| Fluorescence Imaging | Total Aflatoxin | Artificially inoculated kernels from commercial samples | Stationary | Dual-camera multispectral fluorescence | NDFI | Contamination levels were 0.011 to 20 ppb. Screening of contaminated samples demonstrated a high sensitivity (0.987) and high specificity (0.96) at threshold of 20 ppb | [27] |
| Infrared Imaging | Total Aflatoxin | In field inoculation | Stationary | Reflectance and Transmittance spectra | PLS-DA | >95% accuracy for classifying kernels with >100ppb or <10ppb. | [37] |
| Infrared Imaging | Total Aflatoxin, Total Fumonisin | In field inoculation | Stationary | High speed dual-wavelength Reflectance | FWHM | Absorbance at 750 and 1200 nm correctly identify >99% of kernels. 98% accuracy for classifying kernels with >100ppb or uncontaminated. | [38] |
| Infrared Imaging | Fumonisin FB1, FB2 |
Natural contamination form local farmers | - | Fourier transform near infrared spectroscopy | PLS | Coefficients of correlation, root mean square error and standard error of calibration were 0.964, 0.630 and 0.632, respectively | [10] |
| Infrared Imaging | Aflatoxin AFB1 |
Artificial inoculation from commercial samples | Stationary | Short wave infrared hyperspectral imaging | PLS-DA | Yellow, white, and purple corn were scanned. Classification between kernels < 10 ppb and > 1000 ppb was achieved with an accuracy of 97%. | [36] |
| Infrared Imaging | Total Aflatoxin, Fumonisin | Natural contamination from local farmers | In motion | Infrared, Visible, and Ultraviolet Reflectance | LDA, RF, SVM | Skewed distribution of contamination. Spectrometer capable to classify contamination (sensitivity 77%, specificity 83%) and sort at a lower cost. | [9] |
| Infrared Imaging | Aflatoxin AFB1 |
In field inoculation | Stationary | Short wave infrared hyperspectral imaging | PCA, SVM | 11% of the kernels (13/120) were > 2000 ppb. Classification accuracies were 84% and 83% for calibration and validation set, respectively, at thresholds of 20 ppb and 100 ppb. | [32] |
| Infrared Imaging | Aflatoxin AFB1 |
Surface deposition from commercial samples | Stationary | Visible, near-infrared hyperspectral imaging | FDA, PCA | 96%–100% validation accuracy for classification at 5 thresholds: 0, 10, 20, 100, 500 ppb. | [34] |
| Infrared Imaging/Fluorescence | Total Aflatoxin | Wound Inoculation | In motion | Infrared, Visible, and Ultraviolet Reflectance | RF | 86% sensitivity and 97% specificity at a classification threshold of 20 ppb. Spectral data highly skewed. | [1] |
| Infrared Imaging | Total Aflatoxin | Artificial Inoculation | Stationary | Visible, near- infrared reflectance | PLS-DA | 87% accuracy for classification of contaminated kernels at threshold of 20 ppb and 100 ppb. | [17] |
* Acronyms definition: K nearest neighbors (KNN), Principal component analysis (PCA), Least square support vector machine (LS-SVM), Normalized Fluorescence index (NDFI), Partial Least-square principal component analysis (PLS-DA), Full with half maximum (FWHM), Partial least squares (PLS), Range factor (RF), Support vector machine (SVM), Factorial discriminant analysis (FDA).