LAUSR

Abstract. Significance: Pancreatic ductal adenocarcinoma (PDAC) is one of the leading causes of cancer deaths with a best median survival of only 40 to 50 months for localized disease despite multimodal treatment. The standard tissue differentiation method continues to be pathology with histological staining analysis. Microscopic discrimination between inflammatory pancreatitis and malignancies is demanding. Aim: We aim to accurately distinguish native pancreatic tissue using infrared (IR) spectroscopy in a fast and label-free manner. Approach: Twenty cryopreserved human pancreatic tissue samples were collected from surgical resections. In total, more than 980,000 IR spectra were collected and analyzed using a MATLAB package. For differentiation of PDAC, pancreatitis, and normal tissue, a three-class training set for supervised classification was created with 25,000 spectra and the principal component analysis (PCA) score values for each cohort. Cross-validation was performed using the leave-one-out method. Validation of the algorithm was accomplished with 13 independent test samples. Results: Reclassification of the training set and the independent test samples revealed an overall accuracy of more than 90% using a discrimination algorithm. Conclusion: IR spectroscopy in combination with PCA and supervised classification is an efficient analytical method to reliably distinguish between benign and malignant pancreatic tissues. It opens up a wide research field for oncological and surgical applications.