18-09-2012, 11:14 AM
Symmetry-Based Scalable Lossless Compression of 3D Medical Image Data
1Symmetry-Based Scalable.pdf (Size: 2.84 MB / Downloads: 49)
Abstract
We propose a novel symmetry-based technique for
scalable lossless compression of 3D medical image data. The
proposed method employs the 2D integer wavelet transform to
decorrelate the data and an intraband prediction method to
reduce the energy of the sub-bands by exploiting the anatomical
symmetries typically present in structural medical images. A
modified version of the embedded block coder with optimized
truncation (EBCOT), tailored according to the characteristics
of the data, encodes the residual data generated after prediction
to provide resolution and quality scalability. Performance
evaluations on a wide range of real 3D medical images show an
average improvement of 15% in lossless compression ratios when
compared to other state-of-the art lossless compression methods
that also provide resolution and quality scalability including
3D-JPEG2000, JPEG2000, and H.264/AVC intra-coding.
INTRODUCTION
RECENT years have seen 3D medical image acquisitions
becoming a staple in healthcare practice and research
with many imaging modalities, e.g., magnetic resonance
imaging (MRI) and computed tomography (CT), nowadays
considered part of standard healthcare. The availability of such
exquisite noninvasive in vivo high-resolution data have thus
practically revolutionized medicine with 3D medical images
now being an integral part of patients’ records.
As the amount of 3D medical images generated increases,
the storage, management, and access to these large repositories
is becoming increasingly complex. In current practice, picture
archiving and communication systems (PACS), which contain
a collection of specialized networks and computational infrastructure,
are commonly used for storage, retrieval, distribution,
and visualization of medical images. This requires that the underlying
image data be efficiently stored, accessed, and transmitted
over networks of various bandwidth capacities.
2D INTEGER WAVELET TRANSFORM
The basic idea of the 2D discrete wavelet transform is to represent
a 2D signal as a superposition of a wavelet basis [8], [9].
The coefficients of the basis can then be used to reconstruct
the original signal. The 2D discrete wavelet transform gives a
spatial and frequency representation of 2D signals. Each level
of the transform decomposes its input into four spatial frequency
sub-bands denoted as , , , and . The
approximation low-pass sub-band, LL, is a coarser version of
the original signal, while the other sub-bands represent the high
frequency details in the horizontal, vertical and diagonal directions,
respectively. The decomposition is usually iterated on the
approximation low-pass sub-band, which for most natural images
contains most of the energy [9]. The wavelet transform has
many features that make it suitable for our application, such as
representation of an image at different resolutions and packing
of most of the energy in a few wavelet coefficients.
ENTROPY CODING OF RESIDUAL DATA
In this section, we describe the entropy coding of residual data
using a modified version of the EBCOT algorithm, which generates
a bit-stream that is both resolution and quality scalable.
EBCOT is an image compression algorithm for wavelettransformed
images. EBCOT partitions each sub-band in small
blocks of samples, called code-blocks, and generates a separate
scalable bit-stream for each code-block, . The algorithm
is based on context adaptive binary arithmetic coding and
bit-plane coding, and employs four coding passes to code new
information for a single sample in the current bit-plane . The
coding passes are: 1) zero coding (ZC), 2) run-length coding
(RLC), 3) sign coding (SC), and 4) magnitude refinement
(MR). A combination of the ZC and RLC passes encodes
whether or not sample becomes significant in the current
bit-plane . A sample is said to be significant in the current
bit-plane if and only if . The significance of sample
is coded using ten different context models that exploit the
correlation between the significance of sample and that of
its adjacent neighbors. If sample becomes significant in the
current bit-plane , the SC pass encodes the sign information of
sample using five different context models. The MR pass uses
three different context models to encode the value of sample
only if it is already significant in the current bit-plane .
CONCLUSION
We presented a novel wavelet-based scalable lossless compression
method for 3D medical image data. Data decorrelation
is performed by a 2D integer wavelet transform applied
on slices within the medical image volume. The resulting subbands
are then compressed independently by first employing
a block-based intraband prediction method to reduce their energy,
followed by a modified version of the EBCOT algorithm to
achieve resolution and quality scalability. The novelty of our intraband
prediction method is in that it exploits anatomical symmetries
within the structural data captured to predict the value
of the wavelet coefficients on a block-by-block basis.