50: IDENTIFICATION AND CLASSIFICATION OF TUMOR CELLS IN BARRETT’S CARCINOMA PATIENTS BY HYPERSPECTRAL IMAGING (HSI)

Abstract

Abstract Background and aim Digital pathology will revolutionize the discriminate of malignant and non-malignant cells in histologically specimens. Hyperspectral imaging (HSI), a new technology combing imaging with spectroscopy might be beneficial for tumor cell identification. Methods After resection, specimens (n = 95) were fixed in 4% formaldehyde and slices were conducted (3 μm), which were stained with haematoxylin and eosin (HE). Esophageal adenocarcinoma, tumor stroma, squamous epithelium cells and background were recorded by HSI between the visual and near-infrared light (500-1000 nm). A multi-layer perceptron with 2 hidden layers was used for discrimination. Results We were able to record esophageal adenocarcinoma, tumor stroma and squamous epithelium cells by HSI. Differences in the absorbance of squamous epithelium and esophageal adenocarcinoma cells were determined at eosin’s and hematoxilin’s maximal absorption of 524 nm and 630 nm. For all classifiers, the esophageal adenocarcinoma cells, the tumor stroma cells, the squamous epithelium and the background, the intra group variances were quite low. 412,964 spectra of EAC cells, 32,318 spectra of tumor stroma cells and 55,711 spectra of squamous epithelium cells were measured and analyzed for classification. MLP revealed an accuracy of 71% for EAC and stroma cells and 73% for squamous epithelium cells. Conclusion Squamous epithelium and esophageal adenocarcinoma cells show specific spectral alterations due to their HE-staining, when measured by HSI. However, the sensitivities and specificities was quiet high. For implementation in daily routine procedures, the training algorithms need to be validated to foster a semi-automatic decision-making process in histopathological tumor cell identification.