FAST SEGMENTATION AND LOSSY-TO-LOSSLESS COMPRESSION OFDNA MICROARRAY IMAGES Jianping Hua , Zixiang Xiong , Qiang Wu , and Kenneth Castleman Dept of Electrical Engineering, Texas A&M University, College Station, TX 77843 Advanced Digital Imaging Research, LLC., 2450 S. Shore Blvd., #305, League City, TX 77573 ABSTRACT This paper introduces a fast algorithm for microarray image segmentation and an object-based coding technique to com- press the segmented DNA microarray images using modi- fied embedded block coding with optimized truncation (EBCOT) [1]. Microarray images are segmented into foreground and background by a modified Mann-Whitney test-based algo- rithm, which is much faster than Chen et al.’s original one [2]. By extending EBCOT to arbitrarily shaped objects, our scheme can realize lossless coding of foreground and lossy-to-lossless coding of background separately, a feature EBCOT does not offer, and achieve better compression re- sults than other popular coding schemes like LZW, JPEG- 2000 and JPEG-LS. In addition, our scheme offers similar lossy coding performance as JPEG-2000, in which EBCOT is used. 1. INTRODUCTION DNA microarray imaging is a newly developed technology that has attracted enormous interests among researchers in many fields [3] [4]. Contrary to traditional methods, this technology promises to monitor and identify the expression levels of thousands of genes or even the whole genome on one chip simultaneously. The microarray chips are made by dropping the control and test cDNA samples, which are la- beled with different fluorescent dyes, onto orderly arranged spots on glass microslide containing DNA clones of cer- tain expressed sequence tags. Then microarray images are captured under a fluorescence microscope as image set con- taining two channels corresponding to two fluorescent dyes. The DNA clones are printed onto the microslide by auto- matic arrayers, which have different patterns of pins to form pin-arrays. Fig. 1 shows a typical pin-array in one channel of a microarray image set used in our experiments. Because there can be hundreds of target spots in one pin-array, and many pin-arrays on a single microarray chip, observation of thousands of genes using microarray images is highly par- allel and efficient. In the mean time, because microarray images contain huge amount of data saved at a resolution of WORK SUPPORTED IN PART BY THE NIH, NSF, ARO AND ONR. 16 bits per pixel, each image set is typically above 30MB in size, which demands highly efficient compression meth- ods. The current common method for archiving microarray images is to store them in TIFF format after LZW com- pression. However, such an approach does not exploit 2-D correlation of data between pixels and cannot provide grace- ful degradation of image quality for the downstream image analysis. Although a method was recently introduced in [5], the bitstream it generates is only partially progressive. Fig. 1. A typical pin-array (with 84 target spots) out of 24 in one channel of a microarray image set. In this paper, arbitrarily shaped foreground spots in mi- croarray images are first segmented by our modified Mann- Whitney test-based algorithm, which is much faster than Chen et al.’s original one [2]. The shape information is com- pressed by chain code, which costs about 0.05 bits/pixel. We then introduce a new object-based coding scheme by modifying the embedded block coding with optimized trun- cation (EBCOT) algorithm [1], which is now being used in the JPEG-2000 standard. In EBCOT, coefficients of differ- ent subbands are independently coded. Then the bitstreams of all subbands are multiplexed into a layered one. Although EBCOT can give coding priority to some regions of interest in the image, it cannot code them and other parts of the im- age separately. So it’s not an object-based coding method. Here by extending EBCOT to arbitrarily shaped objects, we present a high efficiency object-based coding scheme which can realize lossless compression of the foreground and lossy-to-lossless compression of the background sepa- rately. 2. SEGMENTATION OF MICROARRAY IMAGES In this paper, we use the segmentation algorithm based on Mann-Whitney test introduced by Chen et al. in [2]. Mann-