ID: 284 (Conflict of Interest: K)

Tumorzellidentifikation und –klassifikation mittels hyperspektraler Bildgebung (HSI) bei Patienten mit Barrett-Karzinom

R.Thieme1, H.Köhler2, C.Chalopin2, M.Maktabi2, T.Neumuth2, A.Melzer2, B.Jansen-Winkeln1, I.Gockel1
1Universitätsklinikum Leipzig AöR, Leipzig
2Universität Leipzig, Leipzig


Hyperspectral imaging (HSI) is a novel and innovative imaging technique. The HSI technology thereby combines imaging with spectroscopy and can be used to identify and classify malignant and non-malignant cells by histology.

Material und Methoden

HSI imaging records light between the visual and near-infrared light (500-1000nm). To show the feasibility of this technique in order to discriminate between squamous epithelium and esophageal adenocarcinoma, 45 specimens from patients with esophageal adenocarcinoma were recorded. In 22 of the 45 investigated cases squamous epithelium was visible. The specimens were fixed in formaldehyde, slices were conducted (3µm), and stained with haematoxylin and eosin (HE). K-nearest neighbours (k-NN), a non-parametric supervised classification learning algorithm, was used for discrimination. It classifies unseen data by looking at the class represented by the majority of each data point’s labelled neighbours based on a distance metric.  


We were able to record esophageal adenocarcinoma cells by HSI in all 45 investigated cases. Differences in the absorbance of squamous epithelium and esophageal adenocarcinoma cells were determined between wave lengths of 500 to 700 nm. The intra group variances of the investigated specimens were quite low, for the squamous epithelium as for the esophageal adenocarcinoma cells. The amount of spectra, measured in esophageal adenocarcinoma cells were 333,275 and for squamous epithelium 74,000. Specificity, sensitivity and precision with a k-NN (k = 5) classifier were 0.74, 0.92 and 0.94 for esophageal adenocarcinoma cells.


Squamous epithelium and esophageal adenocarcinoma cells show specific spectral alterations, when measured by HSI. These characteristics could be used to develop a computer-assisted algorithm to discriminate semi-automated for the presence of adenocarcinoma cells in HE stained specimens to foster decision-making support in histopathological diagnosis.