University of Ioannina
Επεξεργασία εικόνων μικροσυστοιχιών με χρήση ευφυών μεθόδων πληροφοριακών συστημάτων
Abstract
dc:descriptionMicroarrays provide a simple way to measure the level of hybridization of known probes of interest with one or more samples under different conditions. Microarray image processing consists of three main stages. The first stage, called spot addressing and gridding, is the procedure for the detection of each spot in the image and the isolation of each spot into a cell. In the second stage, which is known as segmentation stage, each spot of the image is segmented, to separate the signal from the background pixels. Finally, in the third stage the intensity of each spot is extracted and several quantities can be calculated.For the first stage a generalized method for the spot addressing and the gridding of microarray images is introduced, where the spots are structured either in hexagonal or in rectangular grid. Initially the method indentifies the grid of the image. Next, the method utilizes the properties of both the rectangular and the hexagonal grid, to estimate the locations of the non-hybridized spots. This step detects a number of empty spots in each iteration of the proposed Growing Concentric Polygon (GCP) algorithm. The GCP algorithm grows a rectangular or a hexagonal form, for the rectangular structured images or the hexagonal structured images, respectively, detecting a number of spots on the polygon’s contour. For the segmentation stage a novel pixel-by-pixel supervised segmentation method, which is based on classification techniques, is proposed. The method classifies the pixels of the image into three categories using Bayes classifier or Support Vector Machines (SVM). Apart from the signal and the background pixels, the third class includes pixels of artefacts, pixels of the contour of the spot, and pixels of inner holes which exist in donut spots. A set of features from each pixel is used as input for the classification. The proposed method, is advantageous compared to the clustering-based methods, due to the direct characterization of each pixel to the designated category. Otherwise, using clustering techniques different clusters are generated but no distinction exists between them unless a set of rules is applied to separate them. For the evaluation of the method, both rectangular and hexagonal structured images are employed from the Standard Microarray Database, CNV370 beadchip of Illumina, as well as, simulated images generated by the Nykters simulator. The method results in high accuracy in the detection of the spots ranging from 92 – 99% depending on the dataset used. Additionally, high accuracy results are performed in the segmentation stage, where the signal and background pixels are extracted. Pixels from artefacts are excluded from the intensity extraction stage, to provide more reliable values for the levels of hybridization.
Degree
thesis:*- Grantor dc:publisher
- University of Ioannina
- Year dc:date
- 2011
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Giannakeas, Nikolaos
- Γιαννακέας, Νικόλαος
Subjects
dc:subject × 23- Μικροσυστοιχίες
- Επεξεργασία εικόνας
- Ευφυή συστήματα
- Μηχανική μάθηση
- Τοποθέτηση πλέγματος
- Εντοπισμός κηλίδων
- Κατάτμηση
- Microarrays
- Image processing
- Intelligent systems
- Machine learning
- Microarray gridding
- Spot detection
- Spot addressing
- Segmentation
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Ιατρική και Επιστήμες Υγείας
- Άλλες Ιατρικές Επιστήμες
- Natural Sciences
- Computer and Information Sciences
- Medical and Health Sciences
- Other Medical Sciences
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/34764