Refining thyroid function evaluation: a comparative study of preprocessing methods in diffuse reflectance spectroscopy

Abstract

Thyroid dysfunction, comprising conditions such as hyperthyroidism and hypothyroidism, represents a substantial global health challenge, necessitating timely and precise diagnosis for effective therapeutic intervention and patient welfare. Conventional diagnostic modalities often involve invasive procedures, that could cause discomfort and inconvenience for individuals. The non-invasive techniques like diffuse reflectance spectroscopy (DRS) can offer a promising alternative. This study underscores the critical role of preprocessing methods in enhancing the accuracy of thyroid hormone functionality through a non-invasive approach. In the proposed study the spectral data acquired from the DRS setup are subjected to different preprocessing techniques to improve the efficacy of the prediction model. Thirty individuals with thyroid dysfunction were included in the study, and preprocessing methods such as baseline correction, multiplicative scatter correction (MSC), and standard normal variate (SNV), were systematically evaluated. The study highlights that SNV preprocessing outperformed other methods with a root mean square error (RMSE) of 0.005 and an R² of 0.99. In contrast, MSC resulted in an RMSE of 0.87 and an R² of 0.86, while baseline correction showed a RMSE of 0.84 and an unusual R² of 1.09, indicating potential issues. SNV proved to be the most effective technique.