Counting Statistics
Radiation Detect & Measure, summer 2005 ( RSM)
Sources of Error
• Systematic errors
– Consistently get the same error
• Random errors
– Radiation emission and detection are
random processes
• Blunder
– operator error
Radiation Detect & Measure, summer 2005 ( RSM)
Measures of Central Tendency
• Mean
– Average value
• Median
– Middlemost measurement (or value)
– Less affected by outliers
Example: 8, 14, 5, 9, 12
Mean = 9.6
Median = 9
Radiation Detect & Measure, summer 2005 ( RSM)
Measures of Variability
• Variance
– Measure of variability:
• Standard deviation
– Square root of variance
Radiation Detect & Measure, summer 2005 ( RSM)
Statistical Models for random trials
• Binomial Distribution
• Poisson Distribution
– Simplification of binomial distribution with
certain constraints
• Gaussian or Normal Distribution
– Further simplification if average number of
successes is large (e.g., >20)
Radiation Detect & Measure, summer 2005 ( RSM)
Binomial process
• Trial can have only two outcomes
From: The Essential Physics of Medical Imaging (Bushberg, et al) Radiation Detect & Measure, summer 2005 ( RSM)
Binomial probability density function
(PDF)
• N is total number of trials
• p is probability of success
• x is number of successes
From: The Essential Physics of Medical Imaging (Bushberg, et al) Radiation Detect & Measure, summer 2005 ( RSM)
Binomial probability density function
mean and variance
• N is total number of trials
• p is probability of success
• x is mean, σ is standard deviation
If p is very small and a constant then:
Same as Poisson random process.
Radiation Detect & Measure, summer 2005 ( RSM)
Poisson PDF
• Radioactive decay and detection are Poisson
random processes
– Observation time is short compared to the half-life of the
source
• probability of radioactive decays (i.e., p)remains constant
• probability of a given nucleus undergoing decay is small
• Variance
– Variance = mean = pN = x
• Standard deviation
– Standard deviation = √ variance = √ pN = √ x
• Can estimate standard deviation from a single
measurement
Radiation Detect & Measure, summer 2005 ( RSM)
Confidence Intervals
Interval about measurement Probability that mean is within interval (%)
±0.674σ 50.0
±1.0σ 68.3
±1.64σ 90.0
±1.96σ 95.0
±2.58σ 99.0
±3.0σ 99.7
From: Radiation Detection and Measurement (Knoll, GF) Radiation Detect & Measure, summer 2005 ( RSM)
Raphex Question
D70. How many counts must be collected in an instrument with zero background
to obtain an error limit of 1% with a confidence interval of 95%?
A. 1000
B. 3162
C. 10,000
D. 40,000
E. 100,000
Radiation Detect & Measure, summer 2005 ( RSM)
Raphex Answer
D70. How many counts must be collected in an instrument with zero background
to obtain an error limit of 1% with a confidence interval of 95%?
D. A 95% confidence interval means the counts must fall within t wo standard
deviations (SD) of the mean (N). Error limit = 1% = 2 SD/N, but SD = N1/2.
Thus 0.01 = 2(N1/2)/N = 2/ N1/2. Where [0.01]2 = 4/N and N = 40,000.
Radiation Detect & Measure, summer 2005 ( RSM)
Propagation of Error
Description Operation Standard
Deviation
Multiplication of a number with random cx cσ
error by a number without random error
Division of a number with random error x/c σ/c
by a number without random error
Addition of two numbers containing x 1 + x2 √ σ21 + σ22
random errors
Subtraction of two numbers containing x 1 - x2 √ σ21 + σ22
random errors
Radiation Detect & Measure, summer 2005 ( RSM)
Raphex question
G74. A radioactive sample is counted for 1 minute and produces 900 counts.
The background is counted for 10 minutes and produces 100 counts. The net
count rate and net standard deviation are about ____ and ____ counts.
A. 800, 28
B. 800, 30
C. 890, 28
D. 890, 30
E. 899, 30
Radiation Detect & Measure, summer 2005 ( RSM)
Raphex answer
G74. A radioactive sample is counted for 1 minute and produces 900 c ounts.
The background is cou nted for 10 minutes and p roduces 100 c ounts. The net
count rate and net standard deviation are about ____ and ____ counts/min.
D. The net count rate is:
[ ( Ns/ts) - (Nb/tb)] = [(900/1) - (100/10)] = 890.
The net standard deviation, is:
[ ( Ns/t2s) - (Nb/t2b)] 1/2 = [(900) + (1)] = 30.
Radiation Detect & Measure, summer 2005 ( RSM)
