A hybrid framework integrating serum biochemical markers and FTIR spectroscopy with machine learning for early cancer screening.

Early detection of cancer remains a key challenge in clinical diagnosis. This study explored two serum sample-based early cancer screening methods: one based on biochemical parameters and the other based on infrared spectroscopy data, and combined the two methods with machine learning techniques. By statistically analyzing 20 serum biochemical indices from healthy individuals, lung cancer patients, and colorectal cancer patients, and combining them with the SHapley Additive exPlanations algorithm, eight biochemical parameters were finally identified as a biochemical tumor marker panel. The linear discriminant analysis model constructed based on this had a classification accuracy of 89.03 %, a specificity of 92.16 %, and a sensitivity of 81.63 %. In addition, infrared spectroscopy data of healthy, lung cancer, and colorectal cancer samples were used, and after second-order derivative preprocessing, the proposed ReliefF-Successive Projections Algorithm feature selection strategy was used to optimize the classification model performance. Among the multiple classification models, Cubic- Support Vector Machine achieved perfect performance with an accuracy, sensitivity, and specificity of 100 %. These findings demonstrate the value of biochemical analysis and vibrational spectroscopy in early cancer screening and provide useful support for the development of rapid and non-invasive diagnostic tools.