Research On Automatic Analysis Methods Of Microarray Images

Microarray technology is used for the simultaneous identification of thousands of genes in bioinformatics.The image after scanning contains lots of bioinformatics,so miroarray image processing is a critical part of biochip data analysis.The aim of microarray image processing is getting the spots positions,quantizing data of spots shapes and intensities.This thesis studies the automatic processing methods of microarray images,including four parts:skew correction,gridding,image segmentation and extraction.The skew correction is based on Radon transform,but the values are too large of the 16 bits gray images after Radon transform.In the thesis,the spots' boundary is extracted by Sobel,then computing the skew angle by Radon transform.Boundary extraction could reduce the information of the original image,so the calculation is reduced but the skew angle is got correctly.Gridding is based on the profile algorithm to get the spots positions in the image. The traditional profile method for gridding is automatic but may have absent and redundant lines easily.The improved method combines the statistic theory to profile signals,making the gridding without absence and redundancy,even to the images which have many negative spots.Gridding could make sure the spots position,the numbers of row and column,the radius of spots.They are important parameters in the sequent analysis.The improved method are also used to many kinds of miroarray images,including different arrays,different intensities,and getting good results.A local threshold is chosen in image segmentation,and it is computed automatically by Otsu algarithm in every subregion which has only one spot at most. The local threshold makes the microarray image be binary.Because the impurity and noise could infect the segmentation,the open and close operations of morphology are used to the subregion for excluding the noises.Then the spots are identified by template matching to get the spots' regions.Comparing with region growing algorithm,the template matching could get the same shape of the spots,which is propitious to compare the reactive intensities of different spots.The binary image after segmentation is multiplied with original image,so the results only have the information of the spots' regions.But the region information also includes something in the back.So the final part is spots extraction,and the mean of the back should be got at first.They are computed in every local region,being more accurate.The final results is stored in a two dimension array,the size of the array is the same as the microarray.So the array could stand for the spots of different intensities clearly and quantify the biology information.The methods of the thesis are very automatic,without manual operation,and can be used to many kinds of microarray images.The array of the images have no limit,so the methods adapt to the microarray images.