PCA-Polynomial-ELM Model Optimal for Detection of NS1 Adulterated Salivary SERS Spectra

Abstract

Of recent, there has been much interest in the application of Surface Enhance Raman Spectroscopy (SERS) analysis in the detection of diseases such as dengue. Early diagnosis of dengue affords early intervention, greater chance of cure and prevention of mild dengue progressing into life threatening stage. SERS produces, on the interaction of photons from laser beam with saliva samples, a spectral image of its composition here. In the case of dengue fever, Non-Structural Protein 1 (NS1), being its biomarker, is the biochemical fingerprint to be revealed by SERS. NS1 presents in body fluid such as blood and saliva of patients since day one of infection, that makes NS1 a favourite alternative to antibody types of biomarker. However, the concentration of NS1 in saliva is low, yielding a low intensity SERS spectrum. In addition, the spectrum is usually interfered with undesirable noisy features. Extreme Learning Machine (ELM) is a fast algorithm with its strength in data pattern generalization. It has been applied in pattern recognition and machine learning for classification and regression, with encouraging performance. Our work here intends to determine an optimal polynomial-ELM model in classifying SERS spectra of saliva samples adulterated with NS1, amongst the different models subject to three different termination criteria of Principal Component Analysis (PCA). Performance of '100%' is attained for accuracy, sensitivity, specificity and precision, while '1' for kappa, by combining the cumulative percent of total variance (CPV) termination criterion and polynomial-ELM model of power 2 and constant 0.5.