Enhancing lung cancer diagnostics through Raman spectroscopy and machine learning

Abstract This study investigates the use of machine learning to distinguish between lung cancer patients and healthy individuals by analyzing the chemical composition of serum samples via Raman spectroscopy. Sera samples from confirmed lung cancer patients alongside with control samples from healthy individuals were collected. Notable spectral differences were observed at different Raman shifts between the cancerous and healthy samples. Dimensionality reduction was performed using Principal Component Analysis (PCA), and the biochemical variations were analyzed using an advanced ensemble learning method—specifically, the Extreme Gradient Boosting (XGBoost) algorithm. The model's predictions were validated through cross-checking with the K-Nearest Neighbors (KNN) algorithm. The XGBoost model, evaluated through 10-fold cross-validation, outperformed KNN, achieving 97% accuracy, 98% sensitivity, and a precision and specificity of 96%. These results highlight the potential of Raman spectroscopy combined with machine learning as an effective, non-invasive tool for early detection and screening of lung cancer.