Raman spectroscopy in tandem with machine learning – based decision logic methods for characterization and detection of primary precancerous and cancerous cells

Abstract

Early cancer detection improves patient outcomes, but most Raman spectroscopy research has focused on discriminating between normal and malignant cells, ignoring the essential precancerous stage. This study fills that gap by combining Raman spectroscopy with machine learning methods to characterize and categorize normal (primary fibroblast cells from mouse embryos), precancerous (murine fibroblast cell lines (NIH/3T3)), and malignant mouse fibroblast cells transformed by a murine sarcoma virus (MBM-T) as cancerous cells. Key spectral bands associated with malignancy progression were identified using ANOVA-based feature selection, while Log-likelihood estimation decision logic enhanced classification robustness across multiple measurements per cell. The method was 95.8% accurate in classifying normal from cancerous cells, 91% for normal vs. precancerous cells, and 86% for precancerous vs cancerous cells. These results show that Raman spectroscopy has the potential to be a valuable diagnostic tool for early cancer detection, offering insight into carcinogenesis spectrum indications. This study advances Raman-based diagnostics in oncology by strengthening spectrum analysis and classification algorithms.