|
|
|
|
Doctoral Candidate in Image Processing
Master of Electrical Engineering (automatic control)
| Office: | 2138, Department of Information Technology |
| E-mail: | gavrilovic_at_ieee.org |
| Telephone: | +46 18 471 34 73 |
Using spectral/color information to develop novel image processing methods is the topic of my research. After spending a few years in the field, my impression is that color in machine vision is modeled by mimicking limited animal vision (human eyes and brain in particular). My work comprises creating new reduced chromaticity spaces to help machines in distinguishing colors. The methods based on fundamental principles of optical spectroscopy were published and patented in Sweden and the US (link) and collaboration with the Faculty of Medicine established in order to apply the methods.
In fluorescence microscopy, during image acquisition of multiply labeled specimen, two or more of the emission signals can often be physically located in the same area or very near to one another in the final image due to their close proximity within the microscopic structure. This is known as colocalization, a spectral decomposition problem where classical eigenvector-based multivariate analysis is inapplicable. In 2007, spectral angle histograms were born, an innovative tool for data visualization and analysis in spectral microscopy; it decouples spectral from intensity information and helps us address ill-conditioned decomposition problems.
Main publication:
(link) M. Gavrilovic, C. Wählby: "Quantification of Colocalization and Cross-talk based on Spectral Angles", in Journal of Microscopy, vol. 234(3), pp.311-324, May 2009.
Background fluorescence, also known as autofluorescence, and cross-talk are two problems in fluorescence microscopy that stem from similar phenomena, often solved by spectral unmixing. When biological specimens are imaged, the detected signal often contains contributions from fluorescence originating from sources other than the imaged fluorochrome. This fluorescence could either come from the specimen itself (background fluorescence), or from fluorochromes with partly overlapping emission spectra (cross-talk or bleed-through). In order to resolve spectral components at least two distinct wavelength intervals have to be imaged or used for excitation. To perform classical spectral unmixing, spectral signatures of individual compounds, need to be recorded beforehand for all relevant fluorochromes. In situations when spectral signatures are unknown, the method based on spectral angle hsitograms is employed for extraction of parameters for blind linear or nonlinear spectral unmixing. The algorithm also saves time and space for storage of multispectral images as only a few spectral channels need to be recorded and processed.
Main publication:
(link) M. Gavrilovic, C. Wählby: "Suppression of Autofluorescence based on Fuzzy Classification by Spectral Angles", in Proc. of workshop associated with MICCAI: Optical Tissue Image Analysis in Microscopy, Histopathology and Endoscopy, pp.135-146, London, UK, September 2009.
The collaboration with the Molecular Tools group, Science for Life Laboratory, was featured in Cytometry: "Traditionally, researchers in image analysis have initiated collaborations with researchers in medicine and biotechnology once all biochemical methods are developed and, sometimes, even after all images are acquired. To overcome limitations of such an approach, Uppsala University researchers at the Department of Information Technology and the Department of Immunology, Genetics and Pathology teamed up to develop methods for quantification of specific single molecule events, visualized as multicolored signals in fluorescence microscopy. Gavrilovic, Weibrecht and coworkers present a unified framework for image-based evaluation of biochemical methods used in the early stage of specimen preparation, image, acquisition, and finally quantitative image analysis. The authors first describe spectral properties of fluorescence imaging and image analysis procedures, and then use the presented algorithms to evaluate single molecule interaction detection assays".
Publications:
(link) M. Gavrilovic, I. Weibrecht, T. Conze, O. Söderberg, C. Wählby: "Automated Classification of Multi-colored Rolling Circle Products in Dual-channel Wide-field Fluorescence Microscopy", Cytometry Part A, vol. 79A, pp.518-527, June 2011.
(link) I. Weibrecht, M. Gavrilovic, L. Lindbom, U. Landegren, C. Wählby, O. Söderberg: "Visualizing individual sequence-specific protein-DNA interactions in situ", to appear in New Biotechnology
Project funded by the Swedish Research Council in colabpration with Uppsala University Hospital started in January 2010. We are working on replacing subjective diagnosis of prostate cancer with automatic malignancy grading using a combination of tissue staining and spectral image processing. The tissue separation method immune to variations in staining is a problem addressed many times but remained without the right solution. My task is to develop a novel method for tissue separation that outperforms the state-of-the-art algorithms in the field.
Computer Assisted Image Analysis NV1 (spring 2008) (spring 2009) (spring 2010)
Computer Assisted Image Analysis NV2 (fall 2007)
Fuzzy Sets and Fuzzy Techniques (spring 2010)
From April to December 2010, doctoral student representative to the Board of Centre for Image Analysis, SLU and UU
From September 2010 to July 2011, doctoral student representative to the Faculty board for postgraduate studies (Forskarutbildningsnämnd), Faculty of Science and Technology, UU