Tissue-based optical emission spectroscopy meets pathology: a proof of principle study

Source abstract

Intraoperative frozen section diagnostics remains the standard for surgical margin assessment in oncologic surgery but is limited by interobserver variability, restricted availability, and time constraints. Optical emission spectroscopy (OES), a label-free analytical technique, offers a potential alternative by utilizing emission spectra generated during radiofrequency (RF) ablation to distinguish normal from malignant tissues. In this ambispective, multicenter feasibility study, OES spectra were recorded from 137 liver specimens (fresh and cryopreserved) obtained from patients with cholangiocellular carcinoma (CCC) or colorectal liver metastases. Measurements were performed using two spectrometers (Maya 2000Pro and ESA 4000), and machine-learning models, including support vector machines (SVMs) and neural networks (NNs), were trained to classify spectra as normal or tumorous. Spectral analysis revealed consistent differences between normal and malignant liver tissue in emission lines corresponding to corresponding to Mg (279.6-285.2 nm), Na (330.2 nm), Ca (393.4 and 396.9 nm), and Zn (213.5 nm). Using these data, NNs outperformed SVMs across all subgroups, increasing accuracy from 76.9% to 87.8% for metastases and from 76.0% to 88.4% for CCC. Classification performance was consistent between participating cancer centers and unaffected by sample type (fresh vs. cryopreserved). Prediction accuracy remained independent of tumor cell proportion, suggesting that NNs captured information from both tumor cells and the surrounding microenvironment. OES enables rapid, accurate, and reproducible ex vivo differentiation between normal and malignant liver tissue. With further refinement and clinical validation, this approach could complement or, in selected cases, replace intraoperative frozen sections, particularly in time-critical or resource-limited surgical settings.