Enhancing the Diagnostic Evaluation of Thyroid Functionality Using Diffuse Reflectance Spectroscopy and Regression Models

Abstract

Thyroid dysfunction is a prevalent global health concern that necessitates the development of effective and non-invasive screening methods to enable early detection. The application of Diffuse Reflectance Spectroscopy (DRS) in conjunction with preprocessing and predictive models for thyroid dysfunction diagnosis is investigated. The raw spectral data captured from 31 individuals with thyroid dysfunction are subjected to spectral preprocessing techniques like, Standard Normal Variate (SNV), Multiplicative Scatter Correction (MSC), and Baseline Correction. The preprocessed data subjected to regression models like Partial Least Squares Regression (PLSR), Principal Component Regression (PCR), LASSO, Random Forest, Ridge Regression, Gaussian Process Regression (GPR), and Bayesian Regression were employed to analyse the efficacy of the models. The PLSR model in concurrence with SNV outperforms other regression models by achieving an R ² of 0.93, RMSE of 0.29, and MSE of 0.08, indicating low predictive error. The goodness of fit was also evaluated using Pearson's chi-squared test.