Introduction to Bioengineering

Digital Image Basics

Mónica Abella ([email protected])
7.2.H24
Medical image

• Digital medical images are a representation of the spatial

distribution (map) of some physical or chemical property in the
organism-> numbers not a picture!
• To represent this map as a picture we can assign gray levels or
colors to each value

M2.257.15529-48 Introduction to Bioengineering, Digital Image 2

Image display

Look-up table (LUT)

M2.257.15529-48 Introduction to Bioengineering, Digital Image 3
Medical image

• Digital medical images are a representation of the spatial

distribution (map) of some physical or chemical property in the
organism-> numbers not a picture!
• To represent this map as a picture we can assign gray levels or
colors to each value

LUT

M2.257.15529-48 Introduction to Bioengineering, Digital Image 4

Quality parameters

• Spatial resolution:
Size of the smaller detail we
can see in an image

• Contrast resolution:
Ability to distinguish among
structures with similar value

M2.257.15529-48 Introduction to Bioengineering, Digital Image 5

Spatial resolution
Question: How could I measure the spatial resolution of
my imaging system?

M2.257.15529-48 Introduction to Bioengineering, Digital Image 6

Contrast resolution
Question: How could I measure the contrast resolution
of my imaging system?

M2.257.15529-48 Introduction to Bioengineering, Digital Image 7

Analog Image: 2D signal

Voltage
ECG: 1D signal

time

M2.257.15529-48 Introduction to Bioengineering, Digital Image 8

Analog Image: 2D signal
1D signal

MRI: 2D signal
ECG: 1D signal

M2.257.15529-48 Introduction to Bioengineering, Digital Image 9

Analog Image: 2D signal
1D signal

value
distance (mm)

value

distance (mm)
M2.257.15529-48 Introduction to Bioengineering, Digital Image 10
Continuos (analog) – discrete (digital)

M2.257.15529-48 Introduction to Bioengineering, Digital Image 11

Digitization I: Sampling
• Analog 2D signal, f(x,y): infinite points
• We can neither acquire (not ideal acquisition systems) nor
store infinite points (finite disk memory)
• We only take a finite number of samples

The space is divided in small elements: pixels

Sampling

Matrix size: number of pixels in each direction

M2.257.15529-48 Introduction to Bioengineering, Digital Image 12

Digitization I: Sampling
Different matrix size
What about spatial resolution?

32 x 32
64 x 64
128 x 128
256 x 256
 Field of view (FOV):Real size of
the space represented in the
image

M2.257.15529-48 Introduction to Bioengineering, Digital Image 13

 Digitization I: Sampling

Same FOV
Different matrix size
What about spatial resolution?

256 x 256 128 x 128 64 x 64 32 x 32

M2.257.15529-48 Introduction to Bioengineering, Digital Image 14

Digitalization

• We have gone from infinite points to a finite

number of pixels

Sampling

Now, what is the value at each pixel?

M2.257.15529-48 Introduction to Bioengineering, Digital Image 15

Digitalization II: Quantization
• Points in an analog image can have values arbitrarily
precise (it is a continuous distribution)
• We can neither record (limitations in the acquisition) nor
store (limitations in disk space) values with infinite
precision
• We need to round values: The digital images will have a
limited number of possible values

Sampling Quantization

Bit depth: number of bits used for each pixel (defines max.
precision achievable)
M2.257.15529-48 Introduction to Bioengineering, Digital Image 16
Digitalization II: Quantization

255
...
108
107 Real value: 102.86778…
106
105
104
103
102 Quantization
101 value: 103
100

... There is a
0 quantization
error!!

M2.257.15529-48 Introduction to Bioengineering, Digital Image 17

Digitalization II: Quantization

256 8

4 2

M2.257.15529-48 Introduction to Bioengineering, Digital Image 18

Digitalization II: Quantization

• False contouring

M2.257.15529-48 Introduction to Bioengineering, Digital Image 19

Digitalization II: Quantization
• Bit depth: number of bits used for each pixel (defines
max. precision achievable)
• How many values should I store?
– Amount of information: how much will I need?
– Disk space

• Hounsfield Units: -1000 .. 3000

– Its 4000 different values (for HU
differences of 1 HU)
– If I use 8 bits/pixel: 256 different
values
– If I use 12 bits/pixel: 4096 (minimum
bit depth for CT)

M2.257.15529-48 Introduction to Bioengineering, Digital Image 20

Digitization

Digitization

Sampling Quantization

M2.257.15529-48 Introduction to Bioengineering, Digital Image 21

Digital Image

Question: what is a digital image?, what do we actually

have stored in the PC?

M2.257.15529-48 Introduction to Bioengineering, Digital Image 22

Digital Image

M2.257.15529-48 Introduction to Bioengineering, Digital Image 23

Digital Image

M2.257.15529-48 Introduction to Bioengineering, Digital Image 24

Digital Image

9 5
0

M2.257.15529-48 Introduction to Bioengineering, Digital Image 25

Digital Image

Question: what is a digital image?, what do we actually

have stored in the PC?
• List of numbers (really, a list of
bits!): value of the image at each
point
• Header information: how to
arrange those bits? type of data,
endianness, size of the image, …
• Different image formats: tiff, jpeg,
DICOM, raw,…

M2.257.15529-48 Introduction to Bioengineering, Digital Image 26

ImageJ http://imagej.nih.gov/ij/

• Java-based image processing program developed at the

National Institutes of Health
– Public domain
– Designed with an open architecture that provides extensibility via
Java plugins and recordable macros

M2.257.15529-48 Introduction to Bioengineering, Digital Image 27

Image formats
• File->Open->Cell Colony
– Format?
– Header info?
– Matrix size?
– Bit-depth?, number of possible values?

• File->Open->CTslice.dcm
– Format?
– Image->Show Info
 When was this study acquired?
 What is the manufacturer of the scanner?
– Pixel size?
– Bit-depth?, number of possible values?
– Maximum spatial resolution?
– Can we change the visualization colors? Image->Lookup tables

M2.257.15529-48 Introduction to Bioengineering, Digital Image 28

Pseudo-color

We can not distinguish very similar gray levels

M2.257.15529-48 Introduction to Bioengineering, Digital Image 29

Image formats
• Raw data: StudyIopamiro
– Each study consists of a header file (with extension “.ACT”) and a data
file (extension “.ctf”) stored as: Signed of 16 bits.
– Try File->Open
– To open each study we need to used File->Import->Raw:
1. Open the header file (extension “.ACT”) with a text editor (for example
use the “Wordpad” application).
2. Find data type, endianness, and matrix size:
 Number of pixels on the horizontal direction (Proj_size_radial)
 Number of pixel son the vertical direction (Proj_size_axial)
 Number of images (Projections_per_file)
– Number of possible values?
– Open the three studies and explain what you see in each one

M2.257.15529-48 Introduction to Bioengineering, Digital Image 30

 Image formats and profiles
• File->Open->T1 Head
– Matrix size?
– FOV?
– Bit-depth?
– Number of possible values?
– Maximum spatial resolution?
– Image->Stacks->Reslice

• Draw a line in horizontal direction across the image at a height

of 190.5 mm using the tool . Plot the profile with Analyze-
>Plot Profile.
– What do you see?
– What do you think these profiles could be useful for?

M2.257.15529-48 Introduction to Bioengineering, Digital Image 31

Resolution-pixel size
• Take just one slice: Image->Duplicate
• With Image->Scale reduce the image size using 0.25 as scale factor in both
directions X e Y (X scale, Y scale)
– Select the option Create new window, to save the current image naming it
“Copy” (in Title). Make it bigger with the zooming tool
 What is the field of view?
 What is the matrix size?
 What is the pixel size?
 What is the resolution of the new image?

• Increase the size of image “Copy” with Image->Scale selecting a scale

factor of 4 in both directions (X scale, Y scale).
– What is the pixel size?
– Have we lost information compared to the original image?
– What is the resolution of the new image?

M2.257.15529-48 Introduction to Bioengineering, Digital Image 32

Histogram
Pixel value histogram
• The pixel value histogram, h(z), of an image
represents the number of pixels that have a
given value, z

Question: what would be the histogram of the

following image?

Pixel values
0
1
2
3

M2.257.15529-48 Introduction to Bioengineering, Digital Image 34

Pixel value histogram
• The pixel value histogram, h(z), of an image
represents the number of pixels that have a
given value, z

Question: what would be the histogram of the

following image?
Histogram
Pixel values 18
16
0

Number of pixels
14
12
1 10
8

2 6
4
2
3 0
0 1 2 3
Pixel value

M2.257.15529-48 Introduction to Bioengineering, Digital Image 35

Pixel value histogram
• The pixel value histogram, h(z), of an image
represents the number of pixels that have a
given value, z

Number of pixels

Image value

Question: what is the result of integrating the

histogram (adding up all the values)?
M2.257.15529-48 Introduction to Bioengineering, Digital Image 36
 Pixel value histogram

Question: What can

we say about the
Pixel value Pixel value
images that have
each of these
histograms?

Pixel value Pixel value

M2.257.15529-48 Introduction to Bioengineering, Digital Image 37

 Pixel value histogram
Bright
Dark image image

Question: What can

we say about the
Pixel value Pixel value
images that have
each of these
Low
histograms? contrast High
image contrast
image

Pixel value Pixel value

M2.257.15529-48 Introduction to Bioengineering, Digital Image 38

Pixel value histogram
Question: Which histogram
corresponds to each image? 1

a b

c d

M2.257.15529-48 Introduction to Bioengineering, Digital Image 39

Pixel value histogram
Question: Which histogram
corresponds to each image? 1 a

a b

2 b

3 d

c d

4 c

M2.257.15529-48 Introduction to Bioengineering, Digital Image 40

Image display

• Medical images are a map of some physical or chemical

property in the organism, not a picture!
• To represent this map as a picture we can assign gray levels or
colors to each value

LUT

M2.257.15529-48 Introduction to Bioengineering, Digital Image 41

Image display

M2.257.15529-48 Introduction to Bioengineering, Digital Image 42

Image display: Look-up table (LUT)

• The pixel values are addresses to “look up” the brightness

information which is sent to the monitor and used to “paint”
that pixel on the screen

M2.257.15529-48 Introduction to Bioengineering, Digital Image 43

Digitalization II: Quantization
• Bit depth: number of bits used for each pixel
(defines max. precision achievable)
• How many values should I store?
– Amount of information: how much will I need?
– Disk space

• Hounsfield Units: -1000 .. 3000

– Its 4000 different values (for HU
differences of 1 HU)
– If I use 8 bits/pixel: 256 different
values
– If I use 12 bits/pixel: 4096
(minimum bit depth)

M2.257.15529-48 Introduction to Bioengineering, Digital Image 44

Digitalization II: Quantization
• Bit depth: number of bits used for each pixel
(defines max. precision achievable)
• How many values should I store?
– Amount of information: how much will I need?
– Disk space
• Medical images are usually quantified with 8,
12, or 16 bits
– Human eye can not distinguish between more than
~1000 consecutive grey levels (10 bits)
– Conventional screens: 256 (8 bits)

Question: Then, why using more values

than we can distinguish?
M2.257.15529-48 Introduction to Bioengineering, Digital Image 45
Image display

• Digital medical images are a map of some

physical or chemical property in the organism,
numbers not a picture!
• To represent this map as a picture we can assign
gray levels or colors to each value

M2.257.15529-48 Introduction to Bioengineering, Digital Image 46

Linear gray-level LUT: Window width and level

• To select the range we want

to visualize on the screen
– Our eye can only distinguish
~1000 consecutive values
Visualization gray level

• In most medical systems it is

performed just moving the
Pixel value
mouse
M2.257.15529-48 Introduction to Bioengineering, Digital Image 47
Linear gray-level LUT: Window width and level

Soft-tissue window Bone window

M2.257.15529-48 Introduction to Bioengineering, Digital Image 48

Image display
• Open brain.tif
– Basic parameters:
 Matrix size?
 FOV?
 Bit-depth?, how many different values could be found in this image?
– Determine the size of the smallest detail that can be seen in the image if it
was acquired with a scanner of resolution of 0.1 mm
– How many different pixel values could be found in this image?
– Visualize the histogram with Analize->Histogram
• Make a copy of the image with Image->Duplicate (name it “BrainCopy”)
– Select the image “BrainCopy” and adjust its visualization with Image->Adjust->
Window&Level
– Visualize the histograms of the two images “brain.tif and “BrainCopy” with
Analize->Histogram (after selecting the image), what is the difference?

M2.257.15529-48 Introduction to Bioengineering, Digital Image 49

Window and level
• File->Open->CTslice.dcm
– Type?
– Matrix size?
– FOV?
– Bit-depth?, number of possible values?
– Maximum spatial resolution?
– How could you see detail inside the lung or in the bone?
 Playing with Window&Level find the right values of window and level to
see bone and to see the lungs

M2.257.15529-48 Introduction to Bioengineering, Digital Image 50

Window width

If we have the same window level, in which image

the window width is bigger?

M2.257.15529-48 Introduction to Bioengineering, Digital Image 51

Window level

If we have the same window width, in which

image the window level is higher?

M2.257.15529-48 Introduction to Bioengineering, Digital Image 52

Noise / Artifact
Other problems in the images
• Apart from a wrong sampling (loss of spatial resolution) or a
wrong quantization (loss of contrast resolution), images are
degraded by:
– Noise
 Error in pixel value of stochastic nature: when
repeating the experiment, different noise
 Uncertainties in the acquisition process
– Artifacts
 Error in pixel value of systematic nature: if I repeat
the experiment I get the same image
 Not ideal acquisition conditions: misalignments,
“broken” parts, moving of the patient, interfering
materials within the FOV, …

M2.257.15529-48 Introduction to Bioengineering, Digital Image 54

Noise or artifact?

M2.257.15529-48 Introduction to Bioengineering, Digital Image 55