Medical Image CompressionEECE 541 Multimedia Systems

1
Medical Image Compression EECE 541 Multimedia
Systems

Harjot Pooni Ashish Uthama Victor Sanchez
2
What are medical images ?

• Some examples
• MRI / FMRI (Function Magnetic Resonance)

3
Medical Images

• Dynamic 3D Ultrasound
• PET (Positron emission Tomography)
• CT (computerized Tomograhpy)

4
Why compress medical images?

• Growing need for storage
• Efficient data transmission
• Telemedicine
• Tele-radiology applications
• Real time Tele-consultation.
• PACS (Picture archiving and communication systems)

5
Challenges unique to medical images.

• Compression Algorithms
• Lossy / Lossless
• Medical Images should always be stored in
  lossless format.
• Erroneous Diagnostics and its legal implications.

6
Techniques used

• Compression techniques may be classified into
• Lossy
• Lossless
• Moreover, compression algorithms may be applied
  in the spatial domain or frequency domain

7
JPEG 2000 and JPEG-LS

• High compression efficiency
• Lossless color transformations
• Progressive by resolution and quality
• Multiple component images
• ROI coding (static and dynamic)
• Error resilience capabilities
• Object oriented functionalities (coding,
  information, embedding)

8
Drawbacks of JPEG 2000 and JPEG-LS

• Only looks for redundancy in the frame.
• Does not exploit 3D and 4D redundancy
• 3D Redundancy

9
4D Redundancy

• Exploits temporal redundancy

10
Ordering the data to exploit redundancies

• Transform the problem domain Convert 4D data to
  a sequence of 2D data
• Volume
• Time

11
3D-JPEG 2000

• Part 10 JP3D
  
  
• Part 10 is still at the Working Draft stage. It
  is concerned with the coding of three-dimensional
  data, the extension of JPEG 2000 from planar to
  volumetric images
• -http//www.jpeg.org/jpeg2000/j2kpart10.html
• Some commercial vendors have already come out
  with 3D extensions of JPEG 2000
  http//www.aware.com/products/compression/J2K3D.ht
  ml
• Provides guidelines for the use of JPEG 2000 for
  3D data

12
3D-JPEG 2000 The basic approach

• Wavelet transforms

13
3D-JPEG 2000 The basic approach

• Reorder the 4D data by time or volume
• For each set, apply a 1D wavelet transform along
  the z axis
• Apply JPEG 2000 on each transformed slice

14
Drawbacks

• Does not effectively use the redundancy in the
  4th dimension (Temporal redundancy)
• Movement of object between two slices would
  adversely effect performance
• Object motion is significant in medical imaging
• Patient movement
• Organ movement (Heart, Lung)

15
H.264/AVC

• Latest video coding standard ? uses motion
  compensation and estimation.

Source www.vcodex.com
16
Why use H.264?

• Better Intra frame compression
• Medical images have comparatively more uniform
  areas
• Motion estimation and compensation
• Address temporal redundancies
• Multiple frames may be used to predict a single
  frame.
• Better performance
• Different block sizes for motion estimation
  (16x16, 16x8, 8x8)
• Better performance!
• Improved entropy encoder
• Better performance!!

17
Approach One H.264-VOL
Apply H.264/AVC on slices arranged as shown
above Results
Compression Technique Compression ratio
JPEG2000 2.551
JPEG-LS 3.061
3D-JPEG 2000(VOL) 3.151
H.264-VOL 3.891
18
Approach Two H.264-TIME
Apply H.264/AVC on slices arranged as shown
above Results
Compression Technique Compression ratio
JPEG2000 2.551
JPEG-LS 3.061
3D-JPEG2000 (Time) 7.371
H.264-TIME 12.381
19
Best compression performance
H.264 applied across time ? H.264-TIME
20
How to improve compression efficiency?

• Two ideas
• Get the difference between consecutive image
  slices, then use H.264
• Calculate the residual frames, then use H.264
• Main objective reduce the energy content of each
• image slice.

21
Difference between slices
s coded bit-streams
Slice 1
Slice 2
Difference
22
Residual frames
Volume 1
Volume 2
Volume n
H.264 MC entropy coder (CABAC)
Reference slice
Reference slice
Reference slice
Slice 1
Slice 1
Slice 1
Slice s
Slice s
Slice s
s coded bit-streams
H.264 MC
H.264 MC
H.264 MC
Residual 2
Residual 2
Residual s
Residual s
Original slice
Predicted
Residual
23
Results
Compression Technique Improvement Improvement
Compression Technique H.264 Difference H.264 Residual
3D-JPEG2000 100 100
H.264-TIME 20 27
24
Future improvements

• Contextual encoding ? take into account
  characteristics of image

High motion
Low motion
25
Future improvements

• Low motion areas ? lossy
• High motion areas ? lossless

Lossless
Lossy
26
Future improvements

• Encoding using slices (group of macroblocks)
• First slice for high motion areas
• Second slice for low motion areas
• Slices may be encoded at different rates

First slice
Second slice
27
Questions?