Raman Spectroscopy and Artificial Intelligence for Rapid Identification and Classification of Primary Liver Cancer Cells

In our work, we reported a Machine Learning (ML) assisted Raman Spectroscopy (RS) to improve the diagnosis of hepatocellular carcinoma (HCC), the most common form of liver cancer. In the last decade, RS has been widely used in oncology to analyze biological samples and to elucidate the biomolecular mechanisms underlying the transition to the pathological state. To this end, we applied single-cell Raman analysis to uncultured primary tumor and non-tumor cells obtained from resected liver tissues of a patient with HCC. The differential molecular composition of the cell samples was analyzed using RS. It was found that more nucleic acids were present in the nucleus of the cancer cells. In addition, we developed two ML methods to identify key Raman features useful for discriminating the analyzed cell samples based on Linear Discriminant Analysis (LDA): the first method uses Hyper-parameter optimization, and the second uses Principal Component Analysis (PCA-LDA). Despite the high similarity between the Raman spectra of tumor and non-tumor cells, the LDA-based models provide high accuracy in classifying tumor cell spectra (about 90%). To evaluate the predictive power of the developed ML models, we prepared and tested two cells' samples with different percentages of tumor cells and the obtained results were very close to the real values. These results confirm the effectiveness of the approach with the combination of RS and ML models and qualify it as a valid diagnostic tool for the diagnosis of liver cancer.