A Simulated Annealing Approach for Optimal Wavelength Selection in Spectroscopic Analysis of Thyroid Hormones

Diffuse reflectance spectroscopy (DRS) is a highly effective non-invasive technique for assessing thyroid hormone levels by analyzing light interactions with tissue. Its ability to capture structural variations in a non-destructive manner makes it a valuable tool for thyroid diagnostics. The large number of recorded wavelengths in DRS can introduce redundancy and computational inefficiencies, making it essential to identify the most informative spectral bands for accurate predictive modeling. To achieve this, the study employs the simulated annealing (SA) algorithm, which optimizes wavelength selection by iteratively refining the spectral bands, reducing redundancy, and prioritizing wavelengths that exhibit the strongest correlation with thyroid-stimulating hormone (TSH) levels. The selected wavelengths, 860 nm, 890 nm, and 1020 nm, were validated using partial least squares regression (PLSR), confirming their significance in capturing spectral variability relevant to TSH prediction. The evaluation demonstrated the effectiveness of SA-driven wavelength selection, achieving a high coefficient of determination (R² = 0.970) and resulting in a minimal cross validation root mean squared error (RMSECV) of 0.622 at the 176th iteration. These findings highlight the potential of SA in optimizing wavelength selection for DRS-based thyroid assessments, improving model accuracy and stability while ensuring efficient spectral analysis.