PCA-SCG-ANN for Detection of Non-structural Protein 1 from SERS Salivary Spectra

Abstract

With non-structural protein (NS1) being acknowledged as biomarker for Dengue fever, the need to automate detection of NS1 from salivary surface enhanced Raman spectroscopic (SERS) spectra, with claim of sensitivity up to a single molecule thus become eminent. Choice for Principal Component Analysis (PCA) termination criterion and artificial neural network (ANN) topology critically affect the performance and efficiency of PCA-SCG-ANN classifier. This paper aims to explore the effect of number of hidden node for the ANN topology and PCA termination criterion on the performance of the PCA-SCG-ANN classifier for detection of NS1 from SERS spectra of saliva of subjects. The Eigenvalue-One-Criterion (EOC), Cumulative Percentage Variance (CPV) and Scree criteria, integrated with ANN topology containing hidden nodes from 3 to 100 are investigated. Performance of a total of 42 classifier models are examined and compared in terms of accuracy, precision, sensitivity. From experiments, it is found that EOC criterion paired with ANN topology of 13 hidden node outperforms the other models, with a performance of [Accuracy 91%, Precision 94%, Sensitivity 94%, Specificity 96%].