A2 – Endoscopic multispectral sensor system for optimized tissue discrimination

Research Area A: Sensor Development

During the proliferation of tumor cells, the ratios of the tissue components change, which also leads to a change in the optical properties. Since tumor tissue has different absorption coefficients than healthy tissue, it is possible to distinguish them reliably from each other. For spectroscopic investigation, an approach based on hyperspectral imaging will be pursued in this subproject. This already has various medical applications, such as monitoring blood flow, and will now be applied to tissue differentiation in oncology in the visible and near-infrared range.

Efficient Spectroscopic Measurements due to In-situ Applications

A successful tumour resection is defined by its completeness. Nowadays, in most surgeries the tissue is analyzed using the so-called frozen section technique, where the material is cryogenically frozen, cut into thin slices and stained with haematoxylin eosin (HE) for instance. Afterwards, the specimen is investigated using a microscope where spectroscopic measurements can be taken as well (e.g. hyperspectral imaging using FTIR spectrometers). The complete investigation can last up to 30 minutes and the results of the staining are worse and harder to interpret than those of histopathological inspections, in which the tissue slices are embedded in formalin as well. Hence this research training group aims to get rid of those disadvantages tracing a multimodal approach in which several sensor types are being developed and examined. Those sensors shall enable in-situ investigations, which means that the measurements are undertaken within the living body during surgery. The multimodal approach shall increase the reliability of the sensors due to mutual validation.

Main problems using the frozen section technique:

  • investigation time of approximately 30 minutes
  • bad staining results

The approach in this subproject deals with hyperspectral imaging in the visible (VIS) and near-infrared (NIR) spectral range. The tissue is illuminated with polychromatic light and the spectral response of the tissue is detected two-dimensionally. The spectral and spatial information obtained in this way can be used to differentiate and classify different types of tissue on the basis of their spectrum.

Principle of hyperspectral imaging. Left: Hyperspectral data cube, where each plane represents the image of the object in a different spectral channel. Right: By looking at a pixel along the lambda axis, its spectrum is obtained. Different measurement points have different spectra depending on the type of tissue.

Optimized spectra

The results of the first PhD cohort have shown that a highly limited spectrum in the infrared can be sufficient to distinguish malignant from benign tissue. This will now be applied to the VIS/NIR spectral region and further developed towards more potent classification and optimization algorithms.

The use of a highly limited spectrum offers several advantages:

  • better signal-to-noise ratio
  • redction to the essential data
  • shorter evaluation time

The first goal in this subproject is to find the best suited wavelengths to achieve the best distinguishing result. Therefore, the following approaches are examined:

  1. Development of a measurement setup with high spatial and spectral resolution for data generation.
  2. Data analysis using multivariate statistical methods (e.g. principal component analysis)
  3. Use of neural networks

Sensor Implementation into an Endoscope

Another challenge is to develop a miniaturized sensor so that it can be used endoscopically. Various hyperspectral snapshot systems are conceivable for this purpose, and their implementation must be evaluated in the course of the project. A central question will be how the optimized spectra can be realized.


This image shows Andrea Rüdinger

Andrea Rüdinger


PhD Student A2

This image shows Stephan Reichelt

Stephan Reichelt

Prof. Dr.-Ing.

Supervisor A1 and A2

This image shows Tobias Haist

Tobias Haist


Supervisor of Subproject A1

To the top of the page