Segmented Adaptive DPCM for Lossy Compression of Multispectral MR Images
✍ Scribed by Jian-Hong Hu; Yao Wang; Patrick Cahill
- Publisher
- Elsevier Science
- Year
- 1997
- Tongue
- English
- Weight
- 976 KB
- Volume
- 8
- Category
- Article
- ISSN
- 1047-3203
No coin nor oath required. For personal study only.
✦ Synopsis
of the digitally-based (as opposed to film-based) picture archiving and communications systems (PACS). One of This paper reports a multispectral segmented differential pulse coded modulation (MSDPCM) method for well registered the biggest bottlenecks in this process is the extremely multispectral magnetic resonance (MR) images. Given a set of large amount of data associated with medical images genermultispectral MR images, the MSDPCM method first segments ated by various types of imaging modalities [1,2]. This it into statistically distinct regions which by and large corremakes the storage, retrieval, and transmission of raw image spond to different tissue classes. It then finds a suitable linear data very costly and time consuming. Image compression prediction model (LPM) for each class. The LPMs used here are is aimed at reducing the number of bits required to reprethe well known causal autoregressive (AR) and autoregressive sent an image, so as to lower the storage and communicamoving average (ARMA) models. Finally, the MSDPCM tion costs. Two forms of compression are possible: lossless method quantizes the prediction error in each class using a or reversible, and lossy or irreversible. Reversible compresvector quantizer. The original image set is described by the sion methods yield image data identical to the original segmentation map, the model parameters for each class, and image, but cannot achieve very high compression ratios the quantized prediction errors. The MSDPCM method can (typically 2 to 4). Commonly investigated methods for lossproduce very high compression gains, because the specification of the segmentation map and model parameters requires sig-less compression of medical images are differential pulse nificantly fewer bits than that for the original intensity values. coded modulation (DPCM), discrete cosine transform The MSDPCM method using the backward adaptive ARMA (DCT), hierarchical interpolation (HINT), and multiplicamodel has been applied to head MR images with three spectral tive autoregressive (MAR) [3-5]. Irreversible image combands (one T1 weighted and two T2 weighted, 256 ؋ 256 ؋ pression methods for medical images can achieve compres-12 bits/image). In an informal validation, the compressed imsion ratios of 15 or more; however, there is some ages have been evaluated against the originals by three neuroraobservable distortion of original data. The methods for diologists. Images compressed by an average factor of more lossy compression of medical images include the 8 ϫ 8 than 23 have been regarded as acceptable for clinical film block DCT-based JPEG method [6], full-frame DCT and reading. © 1997 Academic Press bit-allocation based method [7, 8], discrete Fourier transform (DFT) or discrete Hadamard transform (DHT) methods [9], wavelet coding [10], combined-transform coding