EMBEDDED WAVELET REGION-BASED CODING METHODS APPLIED TO DIGITAL MAMMOGRAPHY M´ onica Penedo Medicina y Cirug´ ıa Experimental Hospital General Universitario Gregorio Mara˜ n´ on,Spain William A. Pearman Electr., Computer Systems Engineering Dept. Rensselaer Polytechnic Institute, Troy-NY,USA Pablo G. Tahoces, Miguel Souto, Juan J. Vidal Dept. Electr. Comput.;Dept. Radiolog´ ıa Universidad de Santiago de Compostela,Spain ABSTRACT In this paper, we investigate region-based embedded wavelet compression methods and introduce region-based extensions of the Set Partitioning In Hierarchical Trees (SPIHT) and the Set Partitioning Embedded bloCK (SPECK) coding al- gorithms applied to the breast region of digital mammo- grams. We have compared these region-based extensions, called OB-SPIHT and OB-SPECK, against the original SPI- HT and the new standard JPEG 2000 on five digital mam- mograms compressed at rates ranging from 0.1 to 1.0 bits per pixel (bpp). Distortion was evaluated for all images and compression rates by the Peak Signal-to-Noise Ratio (PSNR). A comparison applying SPIHT and JPEG 2000 to the same set of images with the background pixels fixed to zero was also carried out. For digital mammography, region-based compression methods represent an improvement in com- pression efficiency from full-image methods, also providing the possibility of encoding multiple regions of interest. Keywords: Lossy Image Compression, Region-Based Wavelet Transform, Object-Based Coding, Digital Mammog- raphy. 1. INTRODUCTION Medical images are now almost always gathered and stored in digital format for easy archiving, storage and trans- mission, and to allow digital processing to improve diagnos- tic interpretation. A typical mammogram must be digitized at a resolution of about 4000 x 5000 pixels with 50 μm spot size and 12 bits, resulting in approximately 40Mb of digi- tal data. Such high resolution is required in order to detect isolated clusters of microcalcifications that herald an early- stage cancer. However, processing or transmission time of such digital images could be quite long. Also, archiving the amount of data generated in any screening mammography program becomes an expensive and difficult challenge. An efficient lossy compression scheme to reduce digital data without significant degradation of medical image quality is needed. This work was supported in part by the Xunta de Galicia, and Grant TIC2000-0507 from Ministerio de Ciencia y Tecnolog´ ıa (Spain). M. Pene- do is grateful to Rensselaer Polytechnic Institute for residence support. Previous research works have evaluated, with computer aided diagnosis (CAD) systems or observer’s performance studies, lossy compression in digital mammography. To cite one example, Perlmutter et al. evaluated fifty-seven digital mammograms compressed with the Set Partitioning In Hi- erarchical Trees (SPIHT) algorithm [1]. They found no sig- nificant differences between analog and compressed images even at the lowest bit rate tested, 0.15 bpp (80:1 compres- sion ratio). However, the compression method was applied to the least rectangular area containing the breast region. Various techniques have arisen to encode arbitrarily shap- ed regions inside an image at different quality levels accord- ing to their importance or, in our case, diagnostic relevance. One class of such methods is region of interest (ROI) cod- ing, whereby the whole image is transformed and those co- efficients associated to the ROI are coded at higher precision (up to lossless) than the background. Such methods involve encoding of the full image transform. In this study, we have considered region-based methods that transform and encode only the object or objects of interest within the image. Here, the object is the breast region and remainder is the radio- logical background. We have implemented two new region- based coding methods for digital mammography. These im- age compression methods are adaptations of the work of Z. Lu in object-based video coding [2]. In both methods, a border detection technique segment- ed the mammogram into the tissue area and the radiological background. Then, as part of the texture coding method, a Region-Based Discrete Wavelet Transform (RBDWT) was applied to the tissue area to decompose the arbitrary region into wavelet subbands [3]. In one method, an Object-Based extension of the Set Partitioning In Hierarchical Trees (OB- SPIHT) algorithm [4] was used to encode the wavelet co- efficients. For the other method, an Object-Based extension of the Set Partitioned Embedded bloCK (OB-SPECK) coder was used [5]. We compared both coding region-based trans- form techniques to full transform coding via JPEG 2000 and SPIHT. Since the radiological background is unimpor- tant, we also carried out the comparison using the same im- ages with the background set to zero. The coding gains over well-known full-image compression methods, such as SPI- HT and JPEG 2000, are substantial. 0-7803-7750-8/03/$17.00 ©2003 IEEE. ICIP 2003