| Abstract |
DNA microarrays have demonstrated an excellent potential in correlating specific gene expression profiles to specific conditions (e.g., disease) as they allow the concurrent observation of all known genes. Because patterns of gene expression correlate strongly with function, microarrays are providing unprecedented information both on basic research, such as the expression profiles of different tissues and the effect of deletion of specific genes, as well as on applied research, such as human disease, aging, drug and hormone action, mental illness, diet, and many other clinical matters. Microarray experiments, however, involve a large number of error-prone procedures that lead to a high level of noise in the resulting data. The high level of the uncertainty associated with each microarray experiment originates by biological variations (corresponding to real differences between different cell types and tissues) and experimental noise. This uncertainty often obscures some of the important characteristics of the biological processes of interest. More specifically, changes in the measured transcript values in the samples render the clustering of genes into functional groups and the classification of samples difficult. A major challenge in DNA microarray analysis is to eliminate the effect of the noise, which has an additive and a multiplicative component, and recover the gene expression measurements. A number of well-known image processing techniques, including soft and hard thresholding, Bayesian denoising based on Gaussian or Laplacian signal modeling, and multiresolution methods that exploit the correlation between the representation coefficients of adjacent scales have been applied to microarray images by ordinarily assuming the presence of either additive or multiplicative noise. In this dissertation, we propose an image denoising method which accounts for both noise components and makes the microarray spot area more homogeneous and more distinctive from their local background. The proposed approach consists of two stages: one that processes the additive component of the noise and one that processes the multiplicative component. The method first performs a multiresolution decomposition of the image and then accounts for the heavy-tailed statistical behavior of the representation coefficients as well as for their strong statistical dependence across multiple scales. The utility of this framework is validated with real microarray data through visual evaluation and quantitative performance metrics.
|
DNA Microarray Image Enhancement in a MultiResolution Framework
