РАМАН-СПЕКТРОСКОПІЯ ТА МЕТОДИ ШТУЧНОГО ІНТЕЛЕКТУ ДЛЯ СКРИНІНГУ РАКУ РОТОВОЇ ПОРОЖНИНИ

Resume. Oral cancer has a high mortality rate due to late diagnosis. Current screening methods lack sensitivity, highlighting the need for new approaches. Raman spectroscopy enables the analysis of biochemical changes, while artificial intelligence, particularly convolutional neural networks (CNN), enhances spectral data processing and the detection of hidden patterns.Objective. To evaluate the effectiveness of CNN in analyzing Raman spectra of saliva for oral cancer screening and compare it with traditional statistical methods.Results. The analysis of three patient groups (healthy, precancerous, and cancerous) revealed that the most sensitive spectral ranges are 1450-1580 cm⁻¹ and 1580-1700 cm⁻¹. The highest CNN accuracy (90%) was achieved when using the full spectral range (830-2100 cm⁻¹), outperforming traditional statistical methods. Utilizing the entire spectrum allows for the identification of hidden patterns and increases model robustness against noise.Conclusions. CNN demonstrates superior performance in spectral analysis for oral cancer screening compared to statistical methods. Full-spectrum analysis is more informative than examining individual peaks. Future research may explore other machine learning techniques and the integration of spectroscopy into clinical practice.