Label-free multiphoton microscopy and machine learning for recognition of hepatocellular carcinoma.

Complete tumor resection is crucial in oncological liver surgery, and the evaluation of intraoperative resection margins is essential to prove R0 resection. This can be challenging for hepatocellular carcinoma (HCC) due to the heterogeneity of both the tumor and background liver tissue. Label-free multiphoton microscopy (MPM) enables tissue analysis based on endogenous optical signals, and has the potential for intraoperative real-time assessment of resection planes. Matched samples of human HCC and background liver tissue from 76 patients were imaged using a multimodal approach, including coherent anti-Stokes Raman scattering, two-photon autofluorescence, and second harmonic generation. The morphological information contained in each channel was reduced to 17 texture parameters that were used for classification. A neural network model was trained on approximately 25,000 images (35 patients) and used to classify a test set of approximately 27,000 images (38 patients) as well as create maps showing the tumor border (3 patients). Label-free MPM revealed HCC growth patterns as well as steatotic and desmoplastic features. Accurate tumor recognition was achieved on low-lateral-resolution MPM images, mimicking the use of endoscopes. The model achieved a test set correct rate of 97.3% (98.2% for liver and 96.5% for tumor). Analysis of the contribution of the different nonlinear signals to the classification showed that autofluorescence plays a key role in discriminating between neoplastic and non-neoplastic tissue. In conclusion, label-free intraoperative optical histopathology of HCC has the potential to improve tumor resection margins. By implementation in endoscopes, MPM may enable on-site tissue analysis for optimization of tumor identification or characterization of liver tissue.