A recent study in Scientific Reports presents a method for identifying liver tumor subtypes during surgery using fiber-optic attenuated total reflection infrared (ATR-IR) spectroscopy combined with machine learning. This label-free technique aims to provide faster and more objective diagnostics than conventional frozen section analysis, which is time-consuming and relies heavily on visual interpretation.
By offering real-time molecular information, the method has the potential to improve diagnostic accuracy and help surgeons make more informed decisions during procedures.
Image Credit: Jo Panuwat D/Shutterstock.com
Infrared Spectroscopy in Cancer Diagnostics
Infrared (IR) spectroscopy measures how IR light interacts with molecular vibrations in biological tissue, producing distinct spectral patterns that correspond to molecules such as proteins, lipids, carbohydrates, and nucleic acids. Unlike traditional histology, IR spectroscopy does not require staining and allows for non-destructive, objective analysis of tissue composition.
Recent advances in fiber-optic systems now allow IR spectroscopy to be used directly in the operating room. Flexible optical fibers coupled with ATR crystals can analyze freshly excised tissue in real time. Machine learning algorithms enhance the diagnostic process by classifying complex spectral data and distinguishing between tumor types based on their molecular features.
Classifying Liver Tumors During Surgery
Researchers applied fiber-based ATR IR spectroscopy combined with supervised machine learning to differentiate primary liver tumors, including hepatocellular carcinoma (HCC), cholangiocellular carcinoma (CCC), and metastatic liver lesions during surgery.
Liver cancer remains a major global health challenge, with a low 5-year survival rate (22 %) and significant tumor heterogeneity complicating surgical treatment. Accurate intraoperative identification of tumor types and margins is crucial to reduce recurrence and preserve healthy tissue.
They collected liver tissue samples from 69 patients undergoing surgical resection. Freshly excised specimens were comprehensively analyzed using an ATR probe equipped with a germanium crystal and silver halide fibers, connected to a Fourier-transform infrared (FT-IR) spectrometer with a liquid nitrogen-cooled detector. Spectra were acquired from freshly cut tissue surfaces, covering a 900 μm diameter area and taking about one minute.
Data preprocessing focused on the 950-1480 cm-1 range, which contains key biomolecular absorption bands with minimal water interference. A two-step supervised machine learning method was used: genetic algorithms selected optimal spectral regions, and discriminant analysis classified tissue types. FT-IR imaging confirmed that point measurements accurately reflected local tissue composition.
Molecular Features Support Tumor Identification
The data showed consistent molecular differences between normal and tumor tissues. Normal tissue exhibited higher glycogen content, indicated by stronger absorption at 1154 cm-1 and in the 1020-1060 cm-1 range, while tumor tissue demonstrated elevated protein levels and reduced glycogen.
Tumors also presented less compact cellular structures, reflected by weaker phosphate signals at 1080 cm-1. Using these spectral markers, the classification model achieved 89% sensitivity, 92 % specificity, and 90 % accuracy in distinguishing normal from tumor tissue in the test set. Most misclassifications involved HCC, whose spectra closely resembled normal liver, suggesting the need for dedicated models specifically for this subtype.
The approach successfully differentiated tumor subtypes based on distinct spectral fingerprints. HCC maintained higher glycogen levels, reflecting its hepatocyte origin, while CCC showed stronger amide-III bands (~1240 cm-1) and collagen-associated signals (1282 and 1340 cm-1).
Metastases were characterized by distinct protein-related bands at 1162 cm-1 and 1240 cm-1. The supervised classification achieved high accuracy for tumor subtype identification, with no errors in the training set, although a larger cohort validation is needed.
Integrating ATR-IR Spectroscopy in Liver Surgery
This method offers a fast, label-free way to classify tissues during surgery. It supports real-time feedback on tumor type, helping guide resection margins and preserving healthy liver tissue. Compared to frozen section analysis, this approach is quicker and less dependent on subjective judgment.
The technique also provides better molecular resolution than imaging methods like MRI or fluorescence-based tools, which often require labeling agents or prior scans. Its ability to classify tumor subtypes during surgery may also inform decisions about postoperative treatment, including chemotherapy or immunotherapy.
Similar methods have shown usefulness in pancreatic and kidney cancers, suggesting broader potential for intraoperative diagnostics.
Download your PDF copy now!
Future Directions
Fiber-optic ATR-IR spectroscopy combined with machine learning offers a practical method for molecular tissue classification during liver cancer surgery. It addresses some of the limitations of standard diagnostic tools and enables faster, objective assessments in real time. While early results are promising, more research is needed to confirm its reliability across larger, more diverse patient populations.
Next steps include improving probe sterilization, integrating spectroscopy tools into surgical instruments, and refining data analysis methods. These developments could make fiber-optic IR spectroscopy a standard tool in cancer surgeries, helping to improve precision and support personalized treatment strategies.
Journal Reference
Bandzeviciute, R., et al. Algorithm-based intraoperative diagnosis of liver tumors using infrared spectroscopy. Sci Rep 15, 20197 (2025). DOI: 10.1038/s41598-025-06250-z, https://www.nature.com/articles/s41598-025-06250-z
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.