Monitoring fungal infection in maize with high resolution X-ray micro computed tomography

[摘要] ENGLISH ABSTRACT: Maize (Zea mays L.) is an important cereal crop used for human food as well as animal feed. Maize is however vulnerable to contamination by fungi that produce harmful mycotoxins. Fusarium verticillioides is among the most frequently isolated fungus from maize and maize-based products worldwide. Conventional methods for evaluation of fungal infection are destructive in nature and involve tedious sample preparation procedures. X-ray micro computed tomography (X-ray micro CT) was used as a non-destructive technique to monitor the effect of fungal damage on the internal structure of maize kernels infected with F. verticillioides.X-ray images of control and infected kernels were acquired post inoculation using high resolution X-ray micro CT over time. After image acquisition, consecutive two-dimensional (2D) cross sectional images were reconstructed into three dimensional (3D) volumes of the maize kernels. Qualitative results were presented as 2D projection images, and 3D volumes which enable visualisation in different views (top, front and side view). More voids were observed especially in the germ and floury endosperm regions of both the control and infected kernels over time. Quantitative parameters including total volume, mean grey value and total volume of voids were calculated. Total volume and mean grey value increased, while total volume decreased over time in both the control and infected kernels. No significant difference (P ≥ 0.05) was reported between the control and infected for the first four days scanned.Algorithms were developed to extract image textural features from selected 2D images of both the control and infected kernels. First order statistics (mean, standard deviation, kurtosis and skewness) and grey level co-occurrence matrix (GLCM) features were extracted from the side, front and top view of each kernel for the days scanned. The outputs from calculation of these textural features were used as inputs for calculating principal component analysis (PCA) and developing classification models using partial least square discriminant analysis (PLS-DA). Clear separation of the control from the infected was seen on day 8 post inoculation using the first order statistical features. Classification accuracies of 97.22% for control and 55.56% for infected kernels was achieved using the developed PLS-DA model. The GLCM extracted features gave a better classification accuracy of 79.16% for infected kernels with less infected kernels classified as controls compared to first order statistics features.This study demonstrated that, although X-ray micro CT cannot be used as a rapid technique for detection of fungal infection especially during early stages of infection, it allows monitoring of structural changes in the kernel over time, and therefore offer a better understanding of the effect of fungal damage on the microstructure of maize kernel at high resolution.

[发布机构] Stellenbosch University