FISH and automated cell image analysis reduce manual counting fatigue and errors.
Dr. Maria Athelogou, Senior Research Scientist, Definiens AG
Prof. Joachim Diebold, Head of the Institute for Pathology at the Kantonspital Luzern, Switzerland.
Dr. Tamara Manuelian, Solution Engineer Life Sciences, Definiens AG
Dr. Guenter Schmidt, Senior Research Scientist, Definiens AG
Dr. Alexander Schipf, Assistant Physician at the Institute for Pathology at the Kantonspital Luzern, Switzerland.
Prof. Gerd Binnig is the Founder and Head of Research, Definiens AG
Even though targeted cancer medicines are expensive, they are highly effective. Personalized medicine is not just a vision for the future, it is becoming a reality for therapies such as breast cancer. However, in order to identify patients who might benefit from a particular treatment, biomarkers need to be developed.
With research for the discovery and development of biomarkers on the rise, the need for accurate methods for detection, quantification, and interpretation of the effects of biomarkers on tissue is increasing as well. Virtual or digital microscopy is used to acquire image data to describe the effect of biomarkers on biopsy tissue, with thousands of images being acquired in automated workflows. Biomarker discovery platforms and workflows are used in pharmaceutical drug discovery and development processes, producing meaningful, but enormous amounts of complex data that has to be analyzed and interpreted in a timely, reliable, and reproducible way.
Conventionally, clinical experts—such as pathologists—analyze and interpret such data in order to make decisions regarding the diagnosis and therapy of the specific cancer. Image analysis algorithms, methods, and software tools can help interpret biomarker-relevant data.
Figure 1: Definiens image analysis results for Immunohistochemistry (IHC) that measures the HER2 protein present on the surface of the tumor cells. Normal cells are shown in blue. Cells with nuclei that show morphological similarities to cancer cell nuclei are yellow. Cells that show brown stained membrane are purple. These specimens are interpreted on a numeric scale:
The HER2/neu oncogene resides on chromosome 17q, encoding a transmembrane tyrosine kinase growth factor receptor. When breast cancers are present, amplification of the HER2/neu gene, or overexpression of the HER2/neu protein, can be detected. There is correlation between gene amplification and protein overexpression.
HER2 gene amplification assessed by fluorescent in situ hybridization (FISH) may improve the predictive ability of this marker for Herceptin (trastuzumab) therapies. Today, HER2 is assessed by the FISH method when there is an indication of equivocal immunohistochemistry (IHC) results. Figure 1 shows an example of the quantification of IHC HER2/neu breast biopsy by using object- and context-based image analysis software.1 In the case of FISH HER2/neu, the chromosome 17 centromere is marked with a green fluorescent signal, while the HER2 gene is marked with an orange or red fluorescent signal.
In conventional analysis, at least 60 or 100 non-overlapping tumor cell nuclei must be selected and the green and orange/pink signals in each cell nucleus counted. The overall gene to chromosome 17 ratio must be calculated as well. Tumor gene amplification is present when the HER2/neu-to-chromosome 17 ratio is calculated to be greater than 2.2.2 In clinical diagnosis and prognosis support, pathologists calculate this ratio manually by counting the marked chromosomes and genes (so-called “fluorescent signals”) for each nucleus of each of the 100 cells and then calculating the quotient of the gene signals to chromosome signals. A key disadvantage of this method is that due to fatigue and loss of concentration,even an experienced pathologist can overlook some of the fluorescent signals. Only small sections of the biopsy tissue of a tumor can be analyzed, also due to fatigue.
In addition, deciding whether a fluorescent signal represents a single gene or multiple overlapping genes is often subjective. When counting the same slice of a tumor biopsy, different pathologists can come up with different results.
At least three dimensions are examined during the analysis. The pathologist counts signals by changing the focus of the microscope along the z-axis, analyzing the same nuclei in different optical slices to avoid counting mistakes such as overlaps or doubles. Although digital microscopes are able to acquire digital optical slices, there are no corresponding image analysis methods to analyze signals (whether solitary or clustered) or to make calculations relative to associated nuclei, which pathologists do manually when calculating the degree of amplification in breast tumor biopsy tissue.
Traditional image processing techniques focus on pixel processing. An image analysis system from Definiens, the Definiens Cognition Network Technology identifies objects instead of individual pixels and makes inferences about the objects by looking at them in context. This enables levels of aggregation and abstraction that mirror intuitive human understanding and deliver semantic classification based on context.
Single nuclei or clusters of nuclei, solitary fluorescent signals, as well as clustered fluorescence signals in image z-stacks, (such as stacks of optical cuts of biopsy tissue marked with FISH technology) can be analyzed. The degree of amplification can be calculated independently for each single, three-dimensional nucleus automatically. By collating these calculations, more than 100 nuclei can be analyzed automatically.3, 4
When optical slice image data represents fluorescent signals, pixels are grouped to form objects—which form signal compartments—which in turn form nuclei, HER2/neu, and chromosome 17 signals in the nuclei (Figure 2). Algorithms enable automated counting of three-dimensional single signals and three-dimensional clusters of signals, distinguishing between overlapping three-dimensional signals in three-dimensional single nuclei. The quotient of the HER2/neu signals to the chromosome 17 signals in each of the automatically-selected single nuclei is also calculated automatically. The count, distances, and relative positions between nuclei, between signals, and between nuclei and signals can be quantified. This supports more detailed morphological quantification of the effect of FISH biomarkers on cancerous tissue.3, 4
About the Authors
Dr. Athelogou is one of the co-inventors of the Definiens Cognition Network Technology. She has developed computer-based monitoring and diagnostic methods, models, and simulations for complex natural systems.
Prof. Joachim Diebold is globally recognized for his outstanding work in molecular pathology.
Dr. Tamara Manuelian received her doctor’s degree in biology at the University Hamburg.
Dr. Guenter Schmidt received his doctor’s degree in physics at the Technical University Tuebingen and is one of the co-inventors of the Definiens Cognition Network Technology.
Dr. Alexander. Schipf is Assistant Physician at the Institute for Pathology at the Kantonspital Luzern, Switzerland.
Prof. Gerd Binnig is the Founder and Head of Research of Definiens. He was awarded with the Nobel Prize for Physics.
Filed Under: Genomics/Proteomics