2. Why should you care
about image quality?
• Image quality of nuclear medicine cameras
can degrade
• You will be assessing image quality daily/
weekly
• Poor image quality can hurt patients
5. Where does image
noise come from?
• A Patient is injected with 100 Bq of 99Tc.
• How many decays will there be in 1 second?
6. Why does this cause
noise?
• Two adjacent segments of a patient’s
Myocardium each take up 100 kBq of 99Tc.
• only 0.1% of all emitted photons are
detected by the gamma camera.
• Will the same number of counts be gathered
from the two segments?
• How many counts will the gamma camera
record from each segment?
8. Basic statistics:
• The uncertainty on a count is the square
root of the count
• Flip a many200 times
coin
• How 100 heads?
• Expect
• Sqrt(100)=10 10
• Uncertainty is or 90 – 110
• Expect 100±10
11. Absolute and relative
noise
• The absolute uncertainty in a count is
sqrt(N)
• More counts: more absolute uncertainty
• The relative uncertainty is noise÷signal
• Relative uncertainty is 1/sqrt(N)
• More counts: less relative uncertainy
12. 0.2
0.15
Frac
0.1 tion
al un
cert
ainty
0.05
0 100 200 300 400
Number of detected photons
13. How to measure noise?
• Standard Deviation (STDEV) of pixels
• Depends on smoothing
• Depends on Pixel size
17. CNR = 1.6
Contrast Noise
Ratio CNR = 6.5
Not the same as
Image Quality
CNR = 16
18. Using Contrast-Noise
ratio
• CNR alone does not describe image quality
• All other things kept constant, CNR does
describe image quality
• CNR is easy to measure
• Can be used for daily QC
19. How to reduce noise:
more counts!
• Increase injected activity
• Increase exposure time
• Increase detector sensitivity
• Increase collimator throughput
32. Intrinsic Detector Resolution
Gamma ray lands
exactly between
two PMTs
200 optical
photons are
emitted and
detected
How many are detected by the upper PMT?
39. INTEGRAL UNIFORMITY:
For pixels within each area (CFOV and UFOV), the
maximum and the minimum values are to be found
from the smoothed data.
Integral Unif. =100% ((Max - Min) / (Max + Min))
40. DIFFERENTIAL UNIFORMITY:
For pixels within each area (CFOV and UFOV) the
largest difference between any two pixels within a set
of 5 contiguous pixels in a row or column.
Differential Uniformity = + 100% ((Max - Min) / (Max +
Min))
44. The goal of a medical image is to do the
best for the patient.
Patient needs / Image tasks
Tumor detection tumor size
etc.
Why are we doing all this?
defect
localization
Accurate diagnosis
Beneficial action
Healthy, happy patient
45. There are two types of task:
Classification:
group into discreet categories
Healthy or diseased
Stage 1, 2, 3
Estimation:
give continuous number
Tumor uptake
Tumor location (x,y,z)
46. We can put the result of a binary
classification into four categories:
Reality Reality
positive negative
Test True False
positive Positive Positive
Test False True
negative Negative Negative
47. Sensitivity is the fraction of positive
patients that are correctly diagnosed
Reality Reality
positive negative
Test True False
positive Positive Positive
Test False True
negative Negative Negative
What about a contaminated test that classifies all patients as positive?
48. Selectivity is the fraction of healthy
patients that are correctly diagnosed
Reality Reality
positive negative
Test True False
positive Positive Positive
Test False True
negative Negative Negative
What about a defective test that classifies all patients as negative?
50. Results can vary depending on
aggressiveness of tester
Always Positive
ng )
Positive when
ti C
ra O
slight suspicion
pe (R
-O stic
er ri
iev cte
Positive c a
ewhen
R ar
very confident
C h
Never Positive
51. Area Under the Curve (AUC) is a
common measure of test effectiveness