PhD research fellowship in 3-D segmentation and texture analysis of cell nuclei
Goal: To develop High-resolution, high-throughput nuclear analysis as a prognostic marker in human cancer – 3-D segmentation and texture analysis.
Quantification of the chromatin structure in cell nuclei by texture analysis may improve the accuracy of diagnosis and prognosis of human cancer, and may also contribute towards a further understanding of the biological processes involved in carcinogenesis.
The Institute for Cancer Genetics and Informatics, Oslo University Hospital has for 20 years had multidisciplinary cooperation with the Department of Informatics, University of Oslo, and the cooperation is now continued in the strategic research initiative MEDIMA.
A number of tools have been developed in our groups for extraction of textural information from digital images, aimed at early diagnosis and prognosis. An increased insight has been gained on which (sub-visual) chromatin structures that carry such information. New insight has also been gained on the statistical treatment of such data sets, resulting in an altered view on the demands that have to be put on development, testing and comparison of methods.
We are at the forefront in developing methods for segmentation and texture analysis of 2-D images of cell nuclei. Thus, we have established a method for automatic segmentation of cell nuclei from microscopy images of routine histological tissue sections of cancer. And very importantly, we have established a unified approach to extracting a compact set of superior textural features for human cancer prognosis from digital microscopy images of cell nuclei.
3-D nuclei from sections
The aim of the project is to extend our segmentation and texture analysis methods to true 3-D segmentation and 3-D texture analysis of cell nuclei based on optical sections. At present, several different large clinical data sets from paraffin-embedded tissue biopsies or operative lesions from cancer patients treated at the Norwegian Radium Hospital or at cooperating hospitals are available for retrospective studies of prostate cancer, colorectal cancer and mammae cancer.
With this method, we try to combine some of the advantages of tissue sections and monolayer by making sections thick enough (16 µm) such that the section contains whole nuclei. The thickness of the optical section will vary with numerical aperture of the microscope lens and wavelength of light. In order to reproduce whole nuclei with high resolution, many optical sections must be taken from the same nucleus to subsequently put together to a reconstructed three-dimensional nucleus. This requires highly advanced image analysis and computer power, but has the advantage of performing high-throughput, high resolution nuclei analysis (Nucleotyping) in three dimensions. Further more, this will also allow for correct quantitative measures of DNA ploidy in sections.
We have already performed a pilot study where 30 cases of prostate cancer were scanned from 16 tick sections resulting in two hundred 100 nm optical sections. Further, we have developed a cluster consisting of 200 computers and we therefore have computer capacity for 3-D segmentation and texture analysis of data sets typically consisting of 500 cases (patients), 20.000 cell nuclei/case and 200 optical sections through each cell nucleus.
Since 1993 our institute has offered DNA ploidy as a cancer diagnostic and prognostic service. We are currently measuring 1000 clinical cases/year and 2000-3000 cases/year that are included in (research) studies [1,2]. There is a desire from the clinic and the pathologists that DNA ploidy could be performed directly on sections, because it would be easier to integrate the results into the current understanding of the disease. This requires 3-D imaging and analysis, and would be a breakthrough for DNA ploidy measurements.
In the first part of this project, we will complete a survey and comparison of evaluation metrics for segmentation of cell nuclei, and further establish a general and robust method for 2-D segmentation of cell nuclei.
In the second part of the project, the aim is to develop our segmentation and texture analysis methodology further for 3-D segmentation and texture analysis of cell nuclei based on optical sections. The 3-D methods should be generally applicable for cell nuclear segmentation and texture analysis on large clinical data sets from light microscopy. Based on the chromatin texture of cell nuclei, the aim is to establish a few adaptive texture features that consistently give reliable diagnostic and prognostic information, and to gain insight into which image chromatin structures that actually contains this type of information.
Applicants for a project in image analysis must hold a Master’s degree or equivalent in computer science, image processing, signal processing, or applied mathematics/physics. Good knowledge in image analysis or signal processing is preferable in addition to a good background in programming and mathematics or physics. The fellowship is a part of the MEDIMA project and is competing with other MEDIMA activities where the best qualified candidate will be selected.
 G.B. Kristensen, W. Kildahl, V.M. Abeler, J. Kaern, C.G. Trope and H.E. Danielsen, Large-scale genomic instability predicts long-term outcome for women with invasive stage I ovarian cancer, Annals of Oncology. 14 (2003), 1494-1500.
 ] M.E. Pretorius, H. Wæhre, V. M.. Abeler, B. Davidson, L. Vlatkovic, R.A. Lothe, K.-E. Giercksky and H.E. Danielsen, Large scale genomic instability as an additive prognostic marker in early prostate cancer , Cellular Oncology 31 (2009) 251–259.