Diagnosis of intraoperative hepatocellular carcinoma based on FTIR vibration spectroscopy using supervised machine learning

Intraoperative diagnosis plays an essential role in cancer surgery by providing fast and accurate information to clinicians to make a decision. The standard workflow for histopathology based intraoperative diagnosis is generally considered to be time-consuming and labor-intensive. Fourier transform infrared (FTIR) vibration spectroscopy technique has been demonstrated to be a useful tool that yields a molecular fingerprint and provides rapid, nondestructive, high-throughput and clinically relevant diagnostic information. In this study, FTIR spectrometer based on synchrotron radiation was applied to collect the IR spectra of the liver cancer tissues and the adjacent non-cancer tissues of hepatocellular carcinoma (HCC) patients. The FTIR data demonstrated that the ratio of 2959/2926cm-1, 1654/1548cm-1, 1084/1548cm-1 and 1455/1398cm-1 had a significant difference between the two groups, which could serve as indicators to distinguish the liver cancer region from the adjacent non-cancer tissue. Then the supervised machine learning techniques including discriminant analysis coupled with principal component analysis (PCA-DA), support vector machines (SVM) and backpropagation neural networks (BPNN) were applied to classify the spectra data. Finally the performance of these models, such as their precision, sensitivity, specificity and accuracy was assessed, and the results have proved that coupling the FTIR vibration spectroscopy with supervised machine learning techniques could be considered as an accurate and efficient method for the intraoperative diagnosis of HCC.