Classification and prediction of HCC tissues by Raman imaging with identification of fatty acids as potential lipid biomarkers.

PURPOSE: Patients with hepatocellular carcinoma (HCC) can only be treated curatively at early stages and then have a favorable prognosis of this often fatal disease. For this reason, an early detection and diagnostic confirmation are crucial. Raman imaging spectroscopy is a promising technology for high-resolution visualization of the spatial distribution of molecular composition in tissue sections. The aim of this study was to investigate molecular information of liver tissue by Raman imaging for classification and diagnostic prediction.
METHODS: Unstained cryosections of human hepatic tissues (23 patients) were measured by Raman spectroscope in the regions of HCC (n = 12) and fibrosis (n = 17). The acquired data set was used to generate a random forest classification model with 101 iterations of sevenfold cross-validation. The models obtained during cross-validation were also used to predict regions of tumor margin (n = 8) aside from independent testing.
RESULTS: Raman spectra differed between malignant and non-malignant tissue regions. Based on these spectral data, a random forest classification model calculated a prediction accuracy of 86 % (76 % sensitivity and 93 % specificity). The ten most important variables were identified at 2895, 2856, 1439, 1298, 1080, 1063, 1023, 937, 920, and 719 cm(-1).
CONCLUSIONS: In this study, Raman imaging spectroscopy was applied successfully for liver tissue to differentiate, classify, and predict with high accuracy malignant and non-malignant tissue regions. Furthermore, the most important differences were identified as the Raman signature of fatty acids. The demonstrated results highlight the enormous potential which vibrational spectroscopy techniques have for the future diagnostics and prognosis estimation of HCC.