The cell is the origin of all life. However, many diseases and tumors also originate in cells. When cells and tissue show abnormal changes, pathology is required. Pathologists cut wafer-thin slices from tissue samples taken, which they then analyse under the microscope in order to answer the crucial question: is it a benign growth or a malignant tumor? Microscopic analysis is a highly responsible task, as the findings decide the subsequent medical action to be taken and the appropriate treatment to be selected.

Cancer cells glow

Technological progress opens up new opportunities in disease research. Fluorescence microscopy makes use of the fact that some substances in human cells react to excitation light by emitting fluorescence. Through the use of multiple fluorescent labeling, cell components, DNA regions or whole DNA sequences can be identified from the different colours. In breast cancer patients, for instance, doctors are particularly interested in chromosome 17 and HER2 status, which promotes tumor growth. Using multi-color fluorescence in situ hybridization (M-FISH) pathologists can count the number of positively labeled (malignant) cells and color-code them. The cell nucleus, for example, is coloured blue, chromosome 17 green and the HER2/neu receptor red. However, labeling and visualisation is also the greatest challenge in M-FISH microscopy: the color signals of some pixels overlap, cannot be clearly localised and make analysis more difficult.

Spectral separation

The researchers Thomas Arnold and Martin DeBiasio, both specialists in spectral image processing at CTR, the Competence Centre for Advanced Sensor Technologies, have tackled the problem. Arnold: “The M-FISH results are like a cloud with a large number of colour pixels that overlap in places, making it difficult to identify and classify the colours. Pathologists therefore need a great deal of experience and have to be extremely meticulous. We have developed a system that separates the data three dimensionally as it were.” The three-dimensional arrangement of the probes is particularly interesting for the two CTR researchers: “The colour perception of each pixel is also affected by its neighbour. This is why we have broken down all the spectra into their constituent parts. The algorithm deciphers the cloud and separates the pixels,” explains Thomas Arnold.
Not only did they take a new approach in software development, but also adapted the hardware. Martin DeBiasio said: “We used special microscope components. In addition to a highly sensitive camera, we opted for a particular light source that would excite the fluorescent signals better and a tunable filter to record the individual images in a very narrow wavelength range.”

Improved image contrast

The two researchers succeeded in proving that spectral image analysis can improve the image contrast and thus increase the image quality. Their tests showed that 22% of the colour pixels could not be clearly identified in classic color images (RGB). Spectral image analysis enabled them to reduce the percentage to 1.1 %. “Doctors can now identify pixels that they could not previously see with the naked eye. As the images are more accurate and analysis conditions standardised, the technology can help the doctor to make the diagnosis,” said the researchers.

The two multispectral specialists were assisted by Dr Franz G Würz from the Pathology Institute at Klagenfurt regional hospital. “Without the tissue images Dr Würz made available to us and his medical expertise, we would not have been able to carry out our analyses,” added DeBiasio.

Their results have appeared in several scientific publications and the two young researchers have also attended international conferences. In the meantime their work has been progressing. They intend to isolate not just three but five or more colors from the cloud in the future.