Transcript Document

Medical Imaging and Pattern
Recognition
Lecture 3
Image Formats
Oleh Tretiak
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
1
This Lecture
• Digital Images
• Applications of Digital Images
• Image Formats
– General
– Medical
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
2
Introduction
• A digital picture is a picture stored in
binary (bits).
• The picture resides in a digital storage
system as a file.
– A file is a sequence of bytes
– One byte consists of 8 bits
• A picture is a rectangular array of pixels
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
3
Pixels, Grey Values and
Quantization
• Conceptually, a monochrome (black and white) image is a
function
f(x, y), sampled over a two-dimensional grid.
• Each sample value is called a pixel (picture element).
• Conceptually, the function is real-valued and has a
continuous range. This is called the grey value of the
pixel.
• On a computer, it is represented with a finite number of
bits. This is called quantization.
• Most frequently, the digital quantity is interpreted as a
nonnegative integer represented by a byte (8 bits).
0 ≤ v < 28 (256)
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
4
Example
Raw picture format, 256x256, 1 byte per
pixel
Pixel sequence in file
1
2
3
...
256
257
258
259
...
512
513
514
515
...
768
...
...
...
...
...
...
...
65,5
36
65,2 65,2
80
81
[ip:KPI/Notes/Lecture 3] ojt% ls -l lena.raw
-rw-r--r-- 1 ojt staff 65536 27 Sep 20:29 lena.raw
[ip:KPI/Notes/Lecture 3] ojt% hexdump -x lena.raw | more
0000000 6464 6567 6466 6966 6a68 696b 6a6a
0000010 6b6a 6764 6162 5d5a 5d5e 5d63 6b6e
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
6c6b
7c8b
5
Calculations
• In the so-called raw format, the file contains only the
gray values of the pixels.
• Bits/picture = Rows x Columns x bits/pixel
• Bytes/picture = Rows x Columns x bytes/pixel
• Example:
– For the previous slide, 256 rows, 256 columns, 1 byte per
pixel.
– Bytes = 256x256x1 = 65536
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
6
Examples
256x256, 4 bit, 32 kB
256x256, 2 bit, 32 kB
256x256, 1 bit, 8 kB
256x256, 4 bit, 32 kB
128x128, 4 bit, 16 kB
256x256, 8 bit, 65 kB7
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
Pixels, Quantization, and Quality
• A given picture can be represented with
different numbers of pixels and various
numbers of bits per pixel.
• Fewer pixels produces lower quality
• Fewer bits per pixel produces lower quality
• There is a tradeoff between quality and
picture storage requirements.
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
8
Uses of Digital Pictures
• Photography
• Medical Imaging
– X-ray
– CT
– Ultrasound
• Movies
– DVD
– Television
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
9
Why Standard Formats?
• Interoperability
– Image made by Nikon, viewed on
computer made by Apple.
• Advantages of standards
– Competition among vendors (lower prices)
– Creation of markets
– Multiple vendors - product cycle safety
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
10
Useful Data About Photograph
•
•
•
•
•
Size (rows and columns)
Size of print (cm)
Size of subject (cm)
Color/BW
What color?
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
11
TIFF
•
•
•
•
Tagged Image File Format
Proprietary, now owned by Adobe
Many different options (easy to write, hard to read)
File contains information about
– Rows and columns
– How many components (colors, overlays)
– Bits per channel
• Example: lena.raw - 65,536 Bytes, lena.tif - 66,304 B
• Extra storage (768 Bytes) used to store information
about image.
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
12
Image Coding (Compression)
• Why compress?
– Store more pictures in same memory
– Spend less time sending picture over web
• Lossless compression:
– Decompress file and get the same picture, bit - for - bit
– Typically, two-fold compression only for gray value images.
• Lossy compression:
– Decompress and get something similar.
– Any amount of compression is possible.
– Tradeoff between image quality and compression.
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
13
Lossless Compression
• Many ways have been developed
• In practice, Lempel-Ziv (zip, gzip, etc) is
the only one used.
• Get 2-fold compressions over raw
format
• Can be included in TIFF
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
14
JPEG
• JPEG = joint photographic experts
group, standard released in 1992.
• In practice (almost) only lossy image
compression scheme used in practice.
• Standard has many options, only one is
used in practice.
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
15
Example of JPEG compression
Very high quality:
compression = 2.33
Photoshop Image
Very low quality:
compression = 115
Produced by MATLAB
with Quality = 0
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
16
Block Diagram — JPEG DCT
• Image is divided into 8x8 blocks
• The discrete cosine transform (DCT) is
computed of each block.
• The transform values are encoded.
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
17
JPEG DCT Decoder
The values for each block are decoded.
Decoded values are inversely transformed (inverse
DCT), producing 8x8 pixel blocks.
The blocks are assembled into a picture.
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
18
JPEG Files
• In practice (2004)
– DCF - Design rules for the Camera File
system (DCF)
• Can include motion pictures, sound.
– EXIF - Exchangeable image file format for
digital still cameras
• Data about time picture taken, focal length, etc.
• File can be uncompressed (TIFF)
– Compressed data in JPEG format
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
19
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
20
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
21
Medical Image Environment
• Imaging devices, procedure rooms
– X-ray, CT, MRI, Ultrasound, Nuclear
• Patient information system
– Patient history, images,
scheduling/management/billing
– Reliable massive storage devices.
• Reading stations
– Radiologists view images and history, generate
reports.
• All connected through network and storage
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
22
DICOM Standard
• Digital Imaging and Communication in
Medicine
• Ongoing standard activity
• Sponsored by the American College of
Radiology (ACR) and National
Electronics Manufacturers Association
(NEMA)
• 22 workgroups
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
23
Features of DICOM
• Unit of data: Imaging Procedure
– Procedure includes collection of images
with specified goal, and includes
specification of disease, organ, imaging
device, contrast agent, etc.
• The DICOM standard is object-oriented.
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
24
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
25
What are “Medical Images”
• The basic unit is an imaging procedure
– Can consist of several images
•
Data can be “three dimensional”
– Multiple slices, e. g. CT
• Auxiliary non-image data
– Patient history
– Contrast agent
– Date, time, ...
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
26
DICOM and Other Standards
• Image standards in DICOM are from
other sources
– TIFF, JPEG
– Distinguishing features:
• 12, 16 bits per pixel
• Schemes for dealing with three dimensions (not
covered by other standards)
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
27
Ultrasound of Breast Lesion
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
28
Issues
• File size
• Image quality
• Interoperability
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
29
Technology Trends:
Telemedicine, PACS
• Trend toward complex systems
– PACS - picture archiving and communications
• Critical resources
– Expensive imaging devices, procedures
– Medical expertise
• Technological solutions
– Bring patient to scanner
– Bring image to expert
• Digital imaging, databases, networks, standards are
an essential part of the answer
MIPR Lecture 3
Copyright Oleh Tretiak, 2004
30