Optimizing Raman spectra pre-processing and classification for prostate cancer detection in tissue specimens

Prostate cancer is a significant health concern, and novel diagnostic tools are needed to improve its detection and treatment. Raman spectroscopy, a powerful analytical technique, shows promise in non-invasively characterizing biological samples, including cancerous tissues. However, the implementation of Raman spectroscopy in clinics for cancer diagnosis is hindered by challenges such as background fluorescence, noise, and data processing. In this study, we propose a carefully designed pipeline that incorporates various preprocessing techniques and classification algorithms to aid in the accurate detection of cancer in tissue specimens using Raman spectroscopy. Our results demonstrate that the combination of Modpoly background correction with vector normalization yields the best classifier performance, achieving an accuracy of 86% and an F1-score of 91%. We also identify the most important features, including carbohydrate, DNA, RNA, Amide-I, Amide-II, and Amide-III, to differentiate between benign and cancerous prostate tissues. While our evaluation was conducted with limited data, we highlight the need for further studies to establish a more robust Raman spectral data processing pipeline for clinical implementation. This study provides valuable insights into Raman data processing and may pave the way for future applications of Raman spectroscopy in cancer diagnostics at clinics.