1 Lossless DNA Microarray Image Compression Source: Thirty-Seventh Asilomar Conference on Signals, Systems and Computers, Vol. 2, Nov. 2003, pp Authors: N. Faramarzpour, S. Shirani and J. Bondy Speaker: Chia-Chun Wu ( 吳佳駿 ) Date: 2005/05/13
2 Outline 1. Introduction 2. Spiral path 3. Proposed method 4. Experimental results 5. Conclusions 6. Comments
3 1. Introduction Microarray images are usually massive in size. about 30MBytes or more They propose the new concept of spiral path which is an innovative tool for spatial scanning of images
4 2. Spiral path The idea is to convert the 2D structure of an image into a 1D sequence which can scan the image in a highly correlated manner while preserving its spatial continuity It can be used for spatial scanning of any image it is more useful for images with circular, or central behavior
5 2. Spiral path Spiral path (a) spiral sequence (b) and its differential sequence (c) (a) (b) (c)
6 2. Spiral path Table Ⅰ Matrix P for An 18 × 19 Image
7 3. Proposed method Extract individual spots Calculated initial center coordinates Divide the sequences Encode Input image Compressed files No Last spot? Tune the spiral path Yes 16 × 16
8 3.1 Spot extraction where Im[i, j] is the image pixel value.
9 3.1 Spot extraction White lines show how spot sub-images are extracted. spot sub-image (16 x 16)
Spot extraction spot sub-image (16 x 16) m Sub = 16, n Sub =
Spiral path fitting where m Sub and n Sub are the size of extracted spot sub- image.
Spiral path fitting Center X = (302×1+379×2+ … + 284×15+264×16)/ ( … ) =89916/10509= 9 Center y = 97214/10509= 9 (9, 9)
Spiral path fitting Spiral path
Pixel prediction where y i s being their pixel values, r i s being their distances from center and n Neighbor is the number of (y i, r i ) pairs. and use ŷ to predict the intensity of our pixel based on r 0, its distance to center. In (3) we have The linear interpolation function:
Pixel prediction Linear interpolation function for 5 neighbors used to predict intensity of the pixel with distance r 0 from the center
Sequence coding First, we have a residual sequence with the length m Sub ×n Sub -1 for a m Sub ×n Sub spot sub-image. Spot parts and background parts of all spot sub-images of the microarray image are concatenated to form two files. Last, the adaptive Huffman coding is chosen for this application.
Sequence coding Spiral path sequence (a) and prediction residual sequence (b) (a) (b) Spot parts Background parts
Sequence coding Spot part (c) and background part (d) of residual sequence (c) (d)
Experimental results Table Ⅱ Cumulative Compressed Size of Original File (in Bytes) OriginalHeader Spot reg.Background reg. Comp- ressed OriginalCodedOriginalCoded 187,7021,44059,46242,798126,92244,05688,294 Header: spiral path centers, and first pixel intensity values
Experimental results Table Ⅲ Compression Ratio of Our Method Compared to Some Others MethodComp. ratioMethodComp. ratio GIF1.54:1Lossless-41.60:1 ZIP1.67:1Lossless-51.70:1 JPEG :1Lossless-61.69:1 Lossless-11.73:1Lossless-71.79:1 Lossless-21.71:1JPEG-LS2.02:1 Lossless-31.64:1Our2.13:1
21 5. Conclusions This paper proposed a lossless compression algorithm for microarray images. Spiral path and linear neighbor prediction are some of the new concepts proposed in this work.
22 6. Comments 從實驗結果可以明顯的發現, Spot 區域的壓 縮率相較於背景區域而言非常的低,因此可 以針對 Spot 區域找到一個更適合的壓縮方法, 以提昇整體的壓縮率。