National and Kapodistrian University of Athens
Μέθοδοι και αλγόριθμοι επεξεργασίας εικόνας για την ακριβή ανίχνευση πρωτεϊνικών κηλίδων (spots) σε πήκτωμα δισδιάστατης ηλεκτροφόρησης
Abstract
dc:descriptionThe main goal of this doctoral dissertation is the development of new methods for improving the accuracy and efficiency of the spot detection and quantification processes when analyzing 2-DGE images. Image analysis is still considered the bottleneck of a differential expression proteomics analysis workflow due of the large variability in the expression profiles of the spots present on the 2DGE images. There are several commercial software packages that promise automation of the whole spot detection and quantification process, but the hard reality today is that they achieve acceptable performance only after a lot of manual spor editing. A first limitation to be addressed is the lack of sophisticated methods for effective denoising. To this end, a novel methodology was developed, based on the Contourlet Transform. An extensive evaluation of its denoising capabilities has been performed using both synthetic and real gel images. It is demonstrated that it leads to improved SNR, and less distortions, relatively to spatial filtering and wavelet based methods developed recently. Since denoising comes early in the image analysis pipeline, the realized benefits are shown to have a very positive effect on downstream analysis operations. For isolating the image background from areas that include protein spots, a new method was introduced, based on Active Contours without Edges, which is shown to be successful, even in areas with faint and/or saturated spots. Finally, individual spots are located within these areas and their volumes are estimated, using a pipeline of unsupervised machine learning operations: 1D mixture modeling of the intensities histogram is applied first to eliminate background pixels. The surviving pixels are then treated as “molecule generators”, to reveal the distribution of the underlying protein species, manifested as spots on the gel image. This reverse engineering approach constitutes a unique innovation of this work has not been applied in biomedical image analysis before. Finally, by fitting 2D Gaussian models on the generated (by random sampling) “molecules” dataset the individual spots boundaries and their quantities are estimated. It is shown that the developed methodology achieves more precise spot detection and more accurate spot quantification than commercially available software tools. Furthermore, it can be fully automated, not requiring any re-calibration every time a new image is analyzed.
Degree
thesis:*- Grantor dc:publisher
- National and Kapodistrian University of Athens
- Year dc:date
- 2010
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Tsakanika, Panagiotis
- Τσακανίκας, Παναγιώτης
Subjects
dc:subject × 20- Πρωτεομική
- Πρωτεομική ανάλυση υψηλής ρυθμοαπόδοσης
- Δισδιάστατη ηλεκτροφόρηση
- Αποθορυβοποίηση εικόνας
- Ανίχνευση κηλίδων
- Κατάτμηση εικόνων
- Ποσοτικοποίηση κηλίδων
- Μοντελοποίηση κηλίδων
- Proteomics
- High throughput proteomics
- Two-dimensional gel electrophoresis
- Ιmage denoising
- Spot detection
- Image segmentation
- Spot quantification
- Spot modeling
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Natural Sciences
- Computer and Information Sciences
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/22546