Author
 |
Eady, Matthew |
|
Park, Bosoon |
|
Submitted to: Journal of Microscopy
Publication Type: Peer Reviewed Journal Publication Acceptance Date: 11/20/2015 Publication Date: 7/1/2016 Citation: Eady, M.B., Park, B. 2016. Classification of Salmonella Enterica serotypes with selective bands using visible/NIR hyperspectral imaging. Journal of Microscopy. 263(1):10-19. Interpretive Summary: Previously, hyperspectral microscopy has been used to differentiate between species and serotypes of foodborne bacteria, on a cellular level for early and rapid detection methods. Here, the objective is to reduce the number of spectral bands necessary for accurate classification of five Salmonella serotypes. Minimizing the collection of spectral bands to only the necessary wavelengths is critical in reducing the large data processing and storage requirement typically associated with hyperspectral imaging. Initially, 89 bands between 450 – 800 nm were collected. Multivariate data analyses identified important spectral bands. Reduced data sets of 20, 12, 7, and 3 bands were selected. An independent second repetition of hyperspectral microscope images were collected at those bands deemed informative. Multivariate data analyses were also used to classify the five Salmonella serotypes at each of these reduced variable sets. The analyses showed that the second repetition had 100% classification accuracy at each reduced band set, with 12 bands proving more robust, compared to a reduction from 99.5% to 84.5% accuracy in the first set (theorized). This suggests that removing the uninformative spectral bands increased classification accuracy, while reducing the data collection process by 86.5%. Technical Abstract: Optical detection of foodborne bacteria such as Salmonella classifies bacteria by analyzing spectral data, and has potential for rapid detection. In this experiment hyperspectral microscopy is explored as a means for classifying five Salmonella serotypes. Initially, the microscope collects 89 spectral measurements between 450 and 800 nm. Here, the objective was to develop correct classification of five serotypes with optimal spectral bands selected through multivariate data analysis (MVDA), thus reducing the data processing and storage requirement necessary for practical application in the food industry. A digital microscope is equipped with an acousto-optical tunable filter, electron multiplying camera, and metal halide lighting source. Images for each of the five serotypes were collected, and informative bands were identified through a principal component analysis, for four abbreviated spectral ranges containing 3, 7, 12, and 20 spectral bands. The experiment was repeated with an independent repetition and images were collected at each of the reduced variable sets, identified by the first repetition. A support vector machine (SVM) was used to classify serotypes. Results showed that with the first repetition, classification accuracy decreased from 99.5% (89 bands) to 84.5% (3 bands), while the second repetition showed classification accuracies of 100%, possibly due to a reduction in spectral noise. The support vector machine regression (SVMR) was applied with cross-validation, and had R2 calibration and validation values > 0.922. While classification accuracy through SVMC showed that as little as 3 bands were able to classify 100% of the samples, the SVMR shows that the smallest root-mean squared-error values were 0.001 and 0.002 for 20 and 12 bands, respectively, suggesting that the 12 band range collected between 586 and 630 nm is optimal for classifying bacterial serotypes, with only the informative HMI bands selected. |
