Chapter 3, Theory of Operation of HSS Systems

Webmaster's Note:

This page needs some serious editing because some of the headings from Corel Ventura's Desktop Publishing program were erroneously transferred into the HTML version. I can fix this, but not tonight (February 27, 2004 at 9:00 p.m.)

The operating principle of many of the Helgeson internal, external, or waste monitors is firmly based on the unchanging laws of mathematics. Specifically, it is based on the behavior of the Poisson Distribution which describes the frequency with which certain random events will occur. The Helgeson counters are designed so that the user will have complete control of the counter's performance.

Radioactivity is always present in our environment. We call this the "background radioactivity." Therefore, any radiation detector measures this radioactivity. As detector sensitivity increases, the frequency of observing natural background radiation also increases. The background contains both beta and gamma radiations. The beta radiations originate primarily from the daughters of radon and thoron gases. These are naturally occurring radionuclides in the uranium and thorium decay chains. The gamma radiations originate from several naturally occurring sources: radioactive materials such as uranium, thorium and their daughter products and from potassium which is a vital part of the human body. Gamma radiations also come from cosmic rays. All of these are called background. The purpose of almost all of the Helgeson monitors is to measure very low levels of radioactivity, such as those found inside the body, those found on the external surfaces of a person or those found in supposedly "clean" wastes These low levels of radioactivity must be measured in the presence of background radiation. Therefore, we need to know how the measurement of radiation counts will vary with time. We must also state the "action point," or the level of radioactivity requiring notification.

To obtain the best sensitivity, we must work very close to the background. If we are measuring the internal deposition of radioactive materials in a person, we risk telling this individual that he has internal (or external) contamination when, in reality, he is not contaminated. This is called a "false positive" result. Likewise, we risk telling a person that he is free of contamination when, in reality, he is contaminated. This is called a "false negative" result. We call the point at which the counter makes this determination the "Calculated Decision Value" (CDV). If we measure a person or a barrel of supposedly "clean" waste, the exact same methods may be used.

This chapter discusses:

  1. the general theory of the Poisson and Normal (also called the Gaussian) Distributions,
  2. the background,
  3. the action point, and
  4. the CDV, the "Calculated Decision Value."

@MAJOR HEADING = Poisson and <$IGaussian Distribution;plot of>Gaussian Distributions

The <$IPoisson Distribution;demonstration program>Poisson Distribution shows how random events may occur. It is a discrete distribution, meaning that you cannot have a fraction of an event. Either there is an event, or there is not an event. We may know the long-term mean number of events that will occur in a given time interval (which may be a decimal fraction), but we cannot know when the next event will occur. (If the term long-term mean bothers you, substitute long-term average, meaning the average over a long time.) We can state, however, that the variance of the <$IPoisson Distribution;demonstration program>Poisson Distribution is equal to the long-term mean. This fact, that the variance equals the long-term mean, is very important in the design of the "<$IQuicky Counter>Quicky," Model I, "<$IQuicky Counter>Quicky," Model III, "<$IQuicky Counter>Quicky," Model IV, "<$IQuicky Counter>Quicky," Model VI and the various "HECM" instruments. Since the standard deviation is the square root of the variance, then we automatically know the standard deviation of a <$IPoisson Distribution;demonstration program>Poisson Distribution when we know the long-term mean. For example, there are about 640 deaths on our highways during a three day weekend. This means that during any three day weekend there is a 95% probability that there will be from

@LETTER BULLET = 640 - 2*SQRT(640) = 589 deaths

to

@LETTER BULLET = 640 + 2*SQRT(640) = 691 deaths.

(Here is an excellent example of a discrete distribution - you cannot have a fractional death. You either have a death, or you do not.)

The shape of the <$IPoisson Distribution;demonstration program>Poisson Distribution curve depends to a great extent on the long-term mean. If the long-term mean is close to or less than 1.0, then the distribution will be highly skewed to the left. The graph of frequency (Y-axis) versus the number of events (X-axis) will have high values for 0, 1, and maybe 2, but will fall off towards zero very rapidly as shown in Table <$R[C#,TABLE301.TXT]3>-<$R[T#,TABLE301.TXT]1>, on the opposite page, where the mean is 1.0 events. Note that when the long-term mean is 1.000, the probability of observing 0 or 1 are the same, namely, 36.8%.

Columns 1 and 2 are graphed in Figure <$R[C#,POISAT1P.IMG]3>-<$R[F#,POISat1P.IMG]1>, on the opposite page. The X value is displayed on the X-axis and the Probability of Observing X only is displayed on the Y-axis.

If the long-term mean is large, such as 20 or more, then the shape of the distribution is symmetrical around the long-term mean. Look at the figure again. Note that the distribution of data points represented by the "+" look quite symmetrical around a mean of 11. This shows that with values as low as 11, the <$IPoisson Distribution;demonstration program>Poisson Distribution is reasonably well represented by the <$IGaussian Distribution;plot of>Gaussian Distribution.

@BODYTEXTPGBRK = The Normal <$IGaussian Distribution;plot of>Gaussian Distribution represents most of the variation in physical measurements, such as the height of all of the male people living in San Francisco who are between the ages of 30 and 90, the length of a piece of wood or the percent of sugar in 1000 cans of Coca Cola. All of these will be fractional numbers and the graph of these frequencies normally will be a smooth, unbroken symmetrical curve around the average value.

<$&TABLE301.TXT>Because the Poisson and <$IGaussian Distribution;plot of>Gaussian Distributions are so similar for larger numbers, such as 20 or greater, the arithmetic for the <$IGaussian Distribution;plot of>Gaussian Distribution may be used for estimating values of the <$IPoisson Distribution;demonstration program>Poisson Distribution.

For a practical d<$Ifiles;demonstration>emonstration of the shape of the Poisson and <$IGaussian Distribution;plot of>Gaussian Distributions, use the "Poisson" program and choose different values for the <$Ibackground>bac<$Iunits;kilograms>kground and <$Iaction point>action point to see how the shapes are plotted.<$&POISAT1P.IMG>

@MAJOR HEADING = Bac<$Iunits;kilograms>kground and Its Statistical Distribution<$M[Statistical Dist]>

The <$Ibackground>bac<$Iunits;kilograms>kground is a function of the environment around the counter. If the counter is in a well shielded and ventilated room, the <$Ibackground>bac<$Iunits;kilograms>kground should be quite constant. If these conditions do not prevail, variations may occur in the <$Ibackground>bac<$Iunits;kilograms>kground as a function of the time of day, the season, and the temperature. It is important to study the variation of <$Ibackground>bac<$Iunits;kilograms>kground as a function of time so you will know how long a <$Ibackground>bac<$Iunits;kilograms>kground to take and how frequently it should be taken. The longer the bac<$Iunits;kilograms>kground, the lower is the minimum <$Isensitivity;analytical>sensitivity. The following discussion illustrates the influence of <$Itime;counting>counting time on the ability to determine the true bac<$Iunits;kilograms>kground.

Assume that the true <$Ibackground>bac<$Iunits;kilograms>kground counting rate is B counts per second. This value is obtained by dividing the total counts, Cb, obtained in T seconds by the <$Itime;counting>counting time, T:

@ONE EQUATION = <$Eroman B~=~roman C sub roman b over roman T>

The total number of counts, Cb, is at least 20. This means that the Gaussian arithmetic may be used.

We also know that if we make another <$Ibackground>bac<$Iunits;kilograms>kground count of T seconds, we will not obtain exactly the same <$Ianswer>answer due to the random variability of radioactive counting. However, we can make a statement about the average counting rate and our estimate of the range of counting rates which should be observed in successive T-second <$Ibackground>bac<$Iunits;kilograms>kground counts:

@ONE EQUATION = <$Eroman B~=~ { roman C sub roman b ~+-~roman t sub roman p ~*~ sqrt{roman C sub roman b } }over roman T>

Expressing this in terms of the counting rate, B, instead of the total count Cb:

@ONE EQUATION = <$Eroman B~=~ roman B ~+-~roman t sub roman p ~*~ sqrt{roman B / roman T }>

where tp is the value from the <$IGaussian Distribution;plot of>Gaussian Distribution for the probability that counts will lie within a certain percentage of all the counting rates possible. For example, if we wish to say that the range of B is 63% of all the counting rates, then tp = 1.0. If tp = 1.645, then we have a 90% probability that the counts will lie within <$E+- 1.645~*~sqrt { roman B / roman T}> of the average, B..

@MINOR HEADING = Two-Sided Limits

If we make a table, it should look something like this:

@PROBABILITY =

@PROBABILITY = tp Probability, two-sided

@PROBABILITY = 1.0 68.269%

@PROBABILITY = 1.645 90%

@PROBABILITY = 1.96 95% (exactly)

@PROBABILITY = 2.0 95% (approximately)

@PROBABILITY = 2.0 95.45% (exactly)

@PROBABILITY = 2.57586 99%

@PROBABILITY = 3.2906 99.9%

@BODYTEXTPGBRK =

<$&TwoSide.IMG>Figure <$R[C#,TwoSide.IMG]3>-<$R[F#,TwoSide.IMG]2>, above, shows uneven two-sided limits. These are called the two-sided values. If the average counting rate is 100 counts per second, then we can state that for a 1-second count, about 95% of the time the true counting rate lies between:

For 1-second,

@FIRST EQUAT'N = <$E{100~*~1~+-~2~*~sqrt{100~*~1}} over 1~=~{100~+-~2~*~sqrt{100}} over 1~=~{100~+-~2~*~10} over 1 ~=>

@EQUATION = 100. + 20. = 120. counts per second.

@LAST EQUATION = 100. - 20. = 80. counts per second.

The following are ranges for 10-second and 100-second counts:

For 10-seconds,

@FIRST EQUAT'N = <$E{100~*~10~+-~2~*~sqrt{100~*~10}} over 10~=~{1000~+-~2~*~sqrt{1000}} over 10~=~{1000~+-~2~*~31.6} over 10 ~=>

@EQUATION = 100. + 6.32 = 106.32 counts per second.

@LAST EQUATION = 100. - 6.32 = 93.68 counts per second.

For 100-seconds,

@FIRST EQUAT'N = <$E{100~*~100~+-~2~*~sqrt{100~*~100}} over 100~=~{10000~+-~2~*~sqrt{10000}} over 100~=~{10000~+-~2~*~100} over 100 ~=>

@EQUATION = 100. + 2.0 = 102.0 counts per second.

@LAST EQUATION = 100. - 2.0 = 98.0 counts per second.

@BODYTEXTPGBRK = These same equations are re-written using the rate instead of the total counts:

For 1-second,

@FIRST EQUAT'N = <$E100~+-~2~*~sqrt{100~/~1}~=~100~+-~2~*~sqrt{100} =~100~+-~2~*~10 ~=>

@EQUATION = 100. + 20. = 120. counts per second.

@LAST EQUATION = 100. - 20. = 80. counts per second.

For 10-seconds,

@FIRST EQUAT'N = <$E100~+-~2~*~sqrt{100~/~10}~=~100~+-~2~*~sqrt{10}~=~100~+-~2~*~3.16 ~=>

@EQUATION = 100. + 6.32 = 106.32 counts per second.

@LAST EQUATION = 100. - 6.32 = 93.68 counts per second.

For 100-seconds,

@FIRST EQUAT'N = <$E100~+-~2~*~sqrt{100~/~100}~=~100~*~+-~2~*~sqrt{1}~=~100~+-~2~*~1 ~=>

@EQUATION = 100. + 2.0 = 102.0 counts per second.

@LAST EQUATION = 100. - 2.0 = 98.0 counts per second.

Summarizing the meaning of the rate equations, we can say that the standard deviation is equal to the square root of the (Counting Rate divided by the Counting Time):

@ONE EQUATION = <$Eroman StdDev~=~sqrt{ roman Counting roman Rate~/~ roman Counting roman Time}>

Remember, these are the two-sided limits, i.e., we have a symmetrical distribution.

@MINOR HEADING = One-Sided Limits<$&One-Sided Limits>

In our work we are not really interested in two-sided limits. However, we want to know the probability that the <$Ibackground>bac<$Iunits;kilograms>kground will exceed a certain value just by chance. These are called one-sided limits. A table for these would look like the following:

@PROBABILITY =

@PROBABILITY = tp Probability, one-sided

@PROBABILITY = 1.0 84.1345%

@PROBABILITY = 1.281 90%

@PROBABILITY = 1.645 95%

@PROBABILITY = 2.326 99%

@PROBABILITY = 3.0933 99.9%

@PROBABILITY =

Consider these conditions:

@BULLETSPACETX = The <$Ibackground>bac<$Iunits;kilograms>kground is 10 counts per second. A 5,000 Bq 137-cesium <$Isources>source measuring 10 cm by 10 cm is placed three inches above one of the corners of a large p<$Iproportional counters>roportional counter. Two sides of the <$Isources>source are directly above the outer edges of the sensitive area of the p<$Iproportional counters>roportional counter. A <$Isources>source in this position increases the counting rate by 5 counts per second versus 8 counts per second if the <$Isources>source is placed directly above the center of the p<$Iproportional counters>roportional counter. We want to be able to detect 5,000 Bq of 137-cesium contamination in a person during a time of 7 seconds.

@BODYTEXTPGBRK =

Let us see what a one-sided distribution looks like graphically. First consider just the bac<$Iunits;kilograms>kground. This example shows that if we make a 7 second count using a p<$Iproportional counters>roportional counter with an average <$Ibackground>bac<$Iunits;kilograms>kground of 10 counts per second and an <$Iaction point>action point of 5 counts per second above the bac<$Iunits;kilograms>kground, then 90% of the counts would be below 11.71 counts per second and 10% would be above that value. Stated differently, if the average <$Ibackground>b<$Itime;background counting>ac<$Iunits;kilograms>kground counting rate is 10 counts per second and you make a series of 7 second counts, then 90% of the time you will obtain the following:

@ONE EQUATION = <$&UPPERONE.IMG><$E10~+~1.281~*~sqrt{10~/~7}~=~11.53 ~roman counts ~roman per ~roman second.>

Compare this value to the Poisson program value of 11.71 counts per second. Additionally, 95% of the time you will observe counting rates which lie between the values of:

@ONE EQUATION = <$E10~+-~1.96~*~sqrt{10~/~7}~=~10.0~+-~1.20 ~roman counts ~roman per ~roman second.>

@ONE EQUATION = <$E~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~=~8.80 ~roman to ~11.20~ roman counts~ roman per ~roman second.>

@BODYTEXTPGBRK = If counting rates as low as 8.8 counts per second in a 7 second count are observed, such values are considered to be appropriate.

@BODYTEXTPGBRK = <$&LOWERONE.IMG>

@MAJOR HEADING = Bac<$Iunits;kilograms>kground <$Iunits;kilograms><$&Bkg Collection>Collection Methods

There are several methods for collecting bac<$Iunits;kilograms>kground. There is also a statistical approach to allocating a limited amount of time so errors are minimized. We shall start our discussion with this last topic. We are quoting generously from Dr. Price's book "Nuclear Radiation Detection" but the same material may be found in many textbooks on statistics. (Reference <$FPrice, William J., "Nuclear Radiation Detection," p 60, McGraw-Hill Book Company, Inc., 1958>)

@MINOR HEADING = Allocation of Time for Sample Versus Bac<$Iunits;kilograms>kground

If a limited amount of time is available for making both a sample (person, barrel, etc.) count as well as a <$Ibackground>bac<$Iunits;kilograms>kground count, the simple application of statistical principles and differential calculus gives the ratio of <$Itime;counting>counting times. Let us define the <$Ibackground>b<$Itime;background counting>ac<$Iunits;kilograms>kground counting rate, counts per second, as <$Er sub b>, the total counting rate (sample plus bac<$Iunits;kilograms>kground) as <$Er sub T>, the <$Ibackground>b<$Itime;background counting>ac<$Iunits;kilograms>kground <$Itime;counting>counting time as <$Et sub b>, and the total <$Itime;counting>counting time as <$Et sub T>. The standard deviation of the net sample counting rate is

<$Esigma sub s~=~ left ( r sub b over t sub b~+ ~r sub T over t sub T right ) sup 0.5>

By differentiation,

<$E2 sigma sub s ~ d~ sigma sub s~=~-~r sub b over {t sub b ~sup 2}~d t sub b~-~r sub T over {t sub T ~sup 2}~d~t sub T>

Setting <$Ed~sigma sub s~=~0>, the condition for minimum error, and <$Ed~t sub b~+~d ~t sub T~=~0>, the condition for constant time, the result

<$Et sub b over t sub T~=~left (~r sub b over r sub T ~right ) sup 0.5>

is obtained for the optimum use of the <$Itime;counting>counting time. To determine this ratio at the start of the experiment us approximation. Adequate values of the two rates may be determined by short counts.

@MINOR HEADING = Measure the Bac<$Iunits;kilograms>kground for a Fixed Period of Time

When most people think of <$Ibackground>bac<$Iunits;kilograms>kground collection, they think of counting the <$Ibackground>bac<$Iunits;kilograms>kground for a <$Imarkers>fixed period of time. The <$Ibackground>bac<$Iunits;kilograms>kground count usually is made just before or just after the count of the person or sample. This is a completely acceptable method and may be chosen from Chapter <$R[C#,Parameters Chapt]8>, "Examine and/or Change Parameters."

@MINOR HEADING = Measure the Bac<$Iunits;kilograms>kground for a Minimum and Maximum Counting Times

<$Ibackground;mode 2>Choosing a minimum and maximum <$Ibackground>b<$Itime;background counting>ac<$Iunits;kilograms>kground <$Itime;counting>counting time gives somewhat better results, especially if the <$Ibackground>bac<$Iunits;kilograms>kground is <$Ivariable>variable. Since 1974 the older versions of Helgeson software for the "Do-It-Yourself" Lay Down Diagnostic Counter have collected bac<$Iunits;kilograms>kgrounds for 16 minutes. If the operator wanted to <$Icount;starting a count of a person>start a count and the current <$Ibackground>bac<$Iunits;kilograms>kground had been running for less than 8 minutes, the current <$Ibackground>bac<$Iunits;kilograms>kground was abandoned and the previous <$Ibackground>bac<$Iunits;kilograms>kground was used. If the <$Ibackground>bac<$Iunits;kilograms>kground had been collected for 8 minutes or more, the current <$Ibackground>bac<$Iunits;kilograms>kground was saved. The rationale was that for an 8 minute in vivo count, an 8 minute <$Ibackground>bac<$Iunits;kilograms>kground was statistically acceptable; however, a <$Ibackground>b<$Itime;background counting>ac<$Iunits;kilograms>kground <$Itime;counting>counting time of less than 8 minutes was not as statistically valid as the longer one.

@BODYTEXTPGBRK = This same rationale is applicable to other counting systems. When used with any of the "<$IQuicky Counter>Quicky" models, where the <$Itime;subject counting time>subject <$Itime;counting>counting time is two minutes or less, then we recommend that the minimum <$Ibackground>b<$Itime;background counting>ac<$Iunits;kilograms>kground <$Itime;counting>counting time be at least 4 minutes. Longer times are preferable. Referencing the statistical work of Currie (Reference <$FCurrie, L. A., "Limits for Qualitative Detection and Quantitative Determination," Analytical Chemistry, Volume 40, Number 3, pp. 586-693. (1968)>), the <$Iminimum sensitivity;function of background and sample counting times>minimum <$Isensitivity;analytical>sensitivity is a strong function of the <$Ibackground>b<$Itime;background counting>ac<$Iunits;kilograms>kground <$Itime;counting>counting time as can be seen from Figure <$R[C#,MINSENBK.IMG]3>-<$R[F#,MINSENBK.IMG]5>, below.<$&MINSENBK.IMG>

@MINOR HEADING = Measure the Bac<$Iunits;kilograms>kground <$Ibackground;mode 3>Over a Sliding Window

<$&BKGTIME1.IMG>Another good method for ensuring a statistically valid <$Ibackground>bac<$Iunits;kilograms>kground is the practice of a "sliding window." The total <$Ibackground>bac<$Iunits;kilograms>kground time is <$Isliding window background mode;number of sampling intervals>divided into "n" short intervals. If the <$Ibackground>bac<$Iunits;kilograms>kground <$Isliding window background mode;principle>collected within the "nth + 1" interval is within certain limits (limits which are chosen by the Site Health Physicist) then the <$Ibackground>bac<$Iunits;kilograms>kground for this interval is added to the current bac<$Iunits;kilograms>kground, and the <$Ibackground>bac<$Iunits;kilograms>kground from the first interval is dropped. If the <$Ibackground>bac<$Iunits;kilograms>kground from the "nth + 1" interval is outside these limits, a new <$Ibackground>bac<$Iunits;kilograms>kground is started. Figure <$R[C#,BKGTIME1.IMG]3>-<$R[F#,BKGTIME1.IMG]6>, below, shows data where there were 24 intervals of 10 seconds each. The <$Ibackground>bac<$Iunits;kilograms>kground was started at "time zero" and did not have a "constant" 4-sigma error until the 25th interval had been acquired. The figure shows the upper and lower 4.0-sigma limits. The value of 4-sigma was chosen arbitrarily. (This is a value in the Parameters Table and may be changed by the Site Health Physicist.)

In order to demonstrate properly the "sliding window" concept, we deliberately changed the values in the 25th and 60th intervals. Scaled portions of the same graph are shown on the opposite page. Figure <$R[C#,BKGTIME2.IMG]3>-<$R[F#,BKGTIME2.IMG]7> shows the range of 1 to 40 intervals. Note that the lower 4-sigma limit on the 25th counting interval is greater than the upper 4-sigma limit on the running average (compare 63.1 counts per second versus 60.4 counts per second). Normally, this would have started a new bac<$Iunits;kilograms>kground, with the first of the new 24 intervals signaling the start of the new bac<$Iunits;kilograms>kground.

@BODYTEXTPGBRK = Figure <$R[C#,BKGTIME3.IMG]3>-<$R[F#,BKGTIME3.IMG]8> shows the range of 40 to 80 intervals. Note that the upper 4-sigma limit on the 60th counting interval is less than the lower 4-sigma limit on the running average (compare 46.7 counts per second versus 54.4 counts per second). Normally, this would have started a new bac<$Iunits;kilograms>kground, with the first of the new 24 intervals signaling the start of the new bac<$Iunits;kilograms>kground.

@BODYTEXTPGBRK = The Site Health Physicist may elect not to start a new <$Ibackground>bac<$Iunits;kilograms>kground series based on only one count which was outside the limits. Instead he may require that "m" intervals fall outside the limits before a new <$Ibackground>bac<$Iunits;kilograms>kground is started.<$&BKGTIME2.IMG><$&BKGTIME2.BOX> This choice is a strong function of the bac<$Iunits;kilograms>kground: if the <$Ibackground>bac<$Iunits;kilograms>kground is highly <$Ivariable>variable, then it is probably proper to start a new bac<$Iunits;kilograms>kground.<$&BKGTIME3.IMG>

@MAJOR HEADING = <$Iaction point>Action Point

The <$Iaction point>Action Point, A, is defined as the amount of radioactive contamination for which an alarm will sound if a person is truly contaminated. Its value, A, in terms of counts per second is determined by placing a calibration <$Isources>source of the energy of concern at a distance from the detector to simulate actual counting.

In the following example an <$Iaction point>action point, A, of 5 counts per second above the <$Ibackground>bac<$Iunits;kilograms>kground of 10 counts per second (for a total of 15 counts per second), a Type 1 error of 10%, and a Type 2 of 1% where chosen. Less than 1% of the time, a result of less than 11.7 counts per second would be obtained by pure chance. If a 7 second count, counting rates between the limits of:

<$E( ~roman B~+~ roman A~)~+-~2~*~sqrt{(~roman B~+~ roman A)~/~ ~roman T }>

<$E( ~10~+~5~)~-~2~*~sqrt{(~10~+~5)~/~ ~7 }~=~12.07 ~roman counts~/~roman sec roman~~and>

<$E( ~10~+~5~)~+~2~*~sqrt{(~10~+~5)~/~ ~7 }~=~17.93 ~roman counts~/~roman sec .>

will be obtained 95% of the time.

Note that there is a wide distribution of possible counting rates when the true counting rate is 15 counts per second.

Now the questions are:

@BULLET SPACE = Where do we set the limit for calling one result positive and another negative?

@BULLET SPACE = What are our chances of being wrong in either case?

@BODYTEXTPGBRK = The <$Ianswer>answers lead directly to the discussion of the Type 1 and Type 2 errors, the CDV, and the <$Itime;counting>counting time.

@MAJOR HEADING = Calculated Decision Value, Type 1 & 2 <$&Type 1 & 2 Error>Errors, and Counting Time

If the true counting rate is 10 counts per second and the <$Itime;counting>counting time is 7 seconds, 95% of the time we will obtain counting rates as low as:

10 - 2*SQRT(10/7) = 7.6 counts per second

and as high as:

10 + 2*SQRT(10/7) = 12.4 counts per second,

This is just the bac<$Iunits;kilograms>kground. So how do we make a determination as to when a result should be called positive?

This question introduces the terms Type 1 error and Type 2 error. The terms come from the teachings of statistics and are probably more easily understood if we call them the False Positive and False Negative points. To illustrate, let us set the Type 1 error equal to 10%. In simpler terms, there is a 10% risk of calling a result positive when it is truly negative. This means that by random chance one of every ten results will be above our decision value.

Let us also set the Type 2 error, or the risk of failing to call a result positive when it truly is, at only 1%.

Remember that we said that we can use the arithmetic of the <$IGaussian Distribution;plot of>Gaussian Distribution to estimate (very closely, most of the time) the values of the <$IPoisson Distribution;demonstration program>Poisson Distribution when the total number of events is 20 or more. Therefore, using our one-sided t(p) values we can calculate the Calculated Decision Value, CDV.

The equation which gives us the CDV based on the <$Ibackground>bac<$Iunits;kilograms>kground only is:

CDV = (B) + t(p,Type 1)*SQRT(B/T)

We know, however, that we can calculate the CDV based on the <$Iaction point>action point and the Type 2 error. The equation is:

CDV = (B + A) - t(p,Type 2)*SQRT((B+A)/T)

Since both of these equations represent the same number, we may set them equal to each other and eliminate the CDV. From the resulting equation we may calculate the <$Itime;counting>counting time, T.

B + t(p,Type 1)*SQRT(B/T) = B + A - t(p,Type 2)*SQRT((B+A)/T)

Subtracting B from both sides of the equation:

t(p,Type 1)*SQRT(B/T) = A - t(p,Type 2)*SQRT((B+A)/T)

Re-arranging,

t(p,Type 1)*SQRT(B/T) + t(p,Type 2)*SQRT((B+A)/T) = A

t(p,Type 1)*SQRT(B) + t(p,Type 2)*SQRT(B+A) = A * SQRT(T)

and finally,

T = {[t(p,Type 1)*SQRT(B) + t(p,Type 2)*SQRT(B+A)]/A}2

As a test, substitute the numbers from the example to see how closely the <$Itime;counting>counting times compare:

@EQUATION =

@EQUATION = Type 1 error = 10%, therefore t(10%) = 1.281

@EQUATION = Type 2 error = 1%, therefore t( 1%) = 2.326

@EQUATION = T = {[1.281 * SQRT(10) + 2.326 * SQRT(10 + 5)]/5}2

@EQUATION = T = {[1.281 * 3.16228 + 2.326 * 3.87298]/5}2

@EQUATION = T = {[4.05087 + 9.00856]/5}2

@EQUATION = T = {[13.0594]/5}2 = {2.611887}2

@EQUATION = T = 6.86 seconds, which rounds up to 7 seconds.

@EQUATION =

@BODYTEXTPGBRK = This agrees with the Poisson program results.

@MAJOR HEADING = Calculation of the Minimum Counting Time

The <$Ibackground>bac<$Iunits;kilograms>kground is different for each detector. The <$Iaction point>Action Points, Type 1 and Type 2 errors may be different, too. As a result, we must calculate the <$Itime;counting>counting time for each detector and select the longest as the system <$Itime;counting>counting time.

@BODYTEXTPGBRK = <$&COUNTIME.TXT>Table <$R[C#,COUNTIME.TXT]3>-<$R[T#,COUNTIME.TXT]2>, below, shows system <$Itime;counting>counting time calculations with variations to each of the previously-cited parameters are varied. Note that the longest <$Itime;counting>counting time occurs for Detector Number 1. Also note that the lower and upper Calculated Decision Values (CDV) are the same. Normally, there is only one pair where this is true. All of the others are different because of the differences in the various parameters. Figures <$R[C#,CLOSECDV.IMG]3>-<$R[F#,CLOSECDV.IMG]9> and <$R[C#,WIDECDV.IMG]3>-<$R[F#,WIDECDV.IMG]10> show two extremes in the data shown in Table <$R[C#,COUNTIME.TXT]3>-<$R[T#,COUNTIME.TXT]2>, below.

<$&CLOSECDV.IMG>

@BODYTEXTPGBRK = <+><$&WIDECDV.IMG>

@MAJOR HEADING = Rationale for "Gross Counting"

Gross counting is defined as accepting all of the counts between wide energy ranges, such as from 100 keV through 3.0 MeV. No attempt is made to do any energy discrimination within the chosen range.

In order to obtain meaningful information by the gross counting method there are certain requirements.

@MINOR HEADING = Uniformity of Radionuclides

You should be able to document that there is uniformity in the types and relative amounts of the various radionuclides in the samples of concern. For example, if you can show from historical sampling information that there is uniformity, then gross counting without spectroscopy is acceptable.

@MINOR HEADING = Isotopic Composition

@BODYTEXTPGBRK = You must have a good knowledge of the average number of gammas per disintegration from the mixture of radionuclides. Figure <$R[C#,WSTMXTR1.IMG]3>-<$R[F#,WSTMXTR1.IMG]11>, below, shows the nuclides and the concentrations found in some typical "green" wastes at a nuclear power electric generating facility. These same data are shown numerically in Table <$R[C#,WSTMIXT1.TXT]3>-<$R[T#,WSTMIXT1.TXT]3>. Note that c<$I60-cobalt>obalt-60 and cobalt-58 make up almost 75% of the radioactivity found. The next few pages show how to use isotopic composition to calculate the effective number of gamma rays emitted per disintegration of a mixture.<$&WSTMXTR1.IMG>

@BODYTEXTPGBRK = <$&WSTMIXT1.TXT>

@MINOR HEADING = Determining the Average Gammas per Disintegration

Using the data from Table <$R[C#,WSTMIXT1.TXT]3>-<$R[T#,WSTMIXT1.TXT]3>, we shall calculate the average number of gammas emitted per disintegration. We should choose the upper and lower energy levels, rather than just take the data from the table for all energies because the detectors may not be able to measure all of the energies emitted by the radionuclide(s). For example, although the radiochemical analysis of the "green" waste showed cesium-137, barium-137m to be present, it is unlikely that the 31.82, 32.19, and 36.49 keV gamma rays could penetrate the walls of the drum or detector. Therefore, these gamma rays should be excluded when calculating the average number of gammas per disintegration.

The analytical results of the representative sample are given in columns A and B. Column A identifies the radionuclide and Column B gives the concentration in mCi/gram. Column C converts mCi/gram to Becquerels/gram (disintegrations per second per gram) by multiplying Column B by 37,000 Bq (d/s) per mCi. Since we need the concentration in Bq/gram, we will do nothing more with this column.

Columns D, E, and F contain standard disintegration information. The energy in keV, the branching ratio and the gammas per disintegration at the stated energy are taken from "RADDECAY" data. The "RADDECAY" program is in the public domain and is periodically updated by its author. The "RADDECAY" program is a normal part of the "HELGE" software.

Column G may now be calculated by multiplying Column B by Columns E, F and 37,000. The results are the values of gammas per second per gram, abbreviated g/sec-gram for the column header. Column H, the percent of the gammas/sec-gram at a specific energy, is calculated for plotting the graph only.

We are ready now to sum the data in column G (refer to Table <$R[C#,WSTMIXT2.TXT]3>-<$R[T#,WSTMIXT2.TXT]4> on page <$R[C#,WSTMIXT2.TXT]3>-<$R[P#,WSTMIXT2.TXT]23>). This table is actually an <$Iextensions>extension of Table <$R[C#,WSTMIXT1.TXT]3>-<$R[T#,WSTMIXT1.TXT]3>. Look at the first line below the repeated c<$I60-cobalt>obalt-60 data. This line is labeled "Total gammas per second per gram." Column G contains the corresponding value of 0.19356. This number, divided by the total disintegrations per second per gram (0.145197 disintegrations per second per gram), gives us the <$Ianswer>answer we are looking for, the average gammas per disintegration per gram, 1.333117, found on the next line: <$Eroman { 0.19356 over 0.145197~=~1.33117 }>. (That degree of precision is not justifiable but is used to show the differences between using all of the gammas and using a portion, as is explained in the next two para<$Igraphs>graphs

There is a problem with this value, however. It covers all energy ranges, even the 31.82, 32.19, and 36.49 keV gammas from the Cs-137 - Ba-137 decay chain. Depending on how the calibration was performed, these energies could or could not have been included in the determination of the calibration factor, counts per second per Becquerel.

@BODYTEXTPGBRK = Column I contains only those gamma rays which have energies of 100-keV or greater. The sum of column I is given in Table <$R[C#,WSTMIXT2.TXT]3>-<$R[T#,WSTMIXT2.TXT]4> on the row labeled "Total gammas per second per gram" and has a value of 0.19289. This number, divided by the total disintegrations per second per gram, 0.125, gives us the <$Ianswer>answer we are looking for, 1.328, the average gammas per disintegration per gram. Let us compare the two numbers:

@3.7.3 HEADER = gammas/sec-gram gammas/dis-gram

@GAMMAS/SECGR = All gammas 0.19356 1.333

@GAMMAS/SECGR = Gammas within the energy window: 0.19289 1.328

<$&WSTMIXT2.TXT>

We now have all the information necessary to perform gross counting:

Total Activity per gram = 0.145197 Becquerels per gram, or

<$Eroman { Total~nCi~per~gram ~=~ {0.145197~ Bq ~ per ~gram)} over {37~Bq ~ per ~ nCi} ~=~ 0.0039243 ~nCi ~per ~ gram }>

Composite Calibration Factor for a 6-detector system (see Figure <$R[C#,WMCALIB1.IMG]2>-<$R[F#,WMCALIB1.IMG]1>) and Chapter 11:

@3.7.3 HEADER = Factor = 0.0172979 cps per Bq or 0.640022 c/s per nCi.

@BODYTEXTPGBRK =

Using Becquerels the calibration factor is:

<$Eroman {Counting ~rate ~from ~1~ gram ~of ~mixture ~=~ 0.0172979 ~*~ 0.145197~ =~0.0025116~cps~per~gram }>,or using nanoCuries it is:

<$Eroman {Counting ~rate ~from ~1~ gram ~of ~mixture ~=~ 0.0039243 ~*~ 0.640022~ =~0.0025116~cps~per~gram }>.

Assume that the net weight of the barrel of waste is 20 pounds. Therefore, the weight in grams is:

<$Eroman { 20 ~pounds~*~ 453.592~ grams ~per ~pound ~=~ 9,071.54~ grams}>

Therefore, the counting rate above <$Ibackground>bac<$Iunits;kilograms>kground from the total radioactivity in the 20 pound barrel is:

<$Eroman {0.0025116 ~counts~ per~ sec~ per~ gram ~*~ 9,071.85~ grams ~=~22.78~ counts ~per~ second}>.

The <$Icalibration;factors>calibration factors are determined by measuring standard radioactive <$Isources>sources in a uniform distribution within a phantom barrel. The results are expressed in counts per second per Becquerel as their basic units but may be converted to counts per second per nanoCurie for those persons not routinely using the new nomenclature of Becquerels, Seiverts, etc.

@WARNING = Attention:The reader should be fully aware that all of the data in Helgeson programs are stored in units of counts, seconds, Becquerels, centimeters, and kilograms. Other units may be used in screen displays or printed forms, but they will have been converted from the preceeding basic units.

@MINOR HEADING = Total Spectrum Versus Photo<$Ipeak>peak

There is one more justification for using gross counting. When you use the total <$Ispectrum>spectrum you will be using many more counts than just using the counts under a <$Ipeak;photopeak>photo<$Ipeak>peak. This increases the <$Ibackground>b<$Itime;background counting>ac<$Iunits;kilograms>kground counting rate, but it also increases the calibration factor, thereby improving the minimum <$Isensitivity;analytical>sensitivity of the system.

@BODYTEXTPGBRK =

@MAJOR HEADING = Sequential Analysis

@MINOR HEADING = <$Isequential analysis;introduction to>Introduction to Sequential Analysis

While performing Whole Body Counts, following the Three Mile Island incident, Helgeson Scientific Services recognized the need for a Whole Body Counter which was capable of counting large numbers of people in a short period of time - a capability which existing equipment could not accommodate. In the event of future accidents, such a high speed counter could measure a large population segment quickly and frequently. Therefore, Helgeson Scientific Services developed the "<$IQuicky Counter>Quicky" In Vivo Counter which utilizes a statistical technique called Sequential Analysis, as well as other analytical techniques. Since then, other suppliers of whole body counting equipment have recognized the potential of such a machine and have built similar counters.

Several methods for determining the amount of radioactivity in an in vivo count are used by different suppliers of this equipment: Simultaneous Linear Equations, <$Ianalysis;least squares>Least Squares Curve Fitting, Peak Search, and Sequential Analysis. The first three methods are widely known and will not be discussed in this paper. The principles of Sequential Analysis will be developed with sufficient detail for the reader to understand its advantages.

"When a series of observations are obtained one after another, as is common in chemical and physical research, it is generally possible to adopt an alternative procedure in which, after each observation is made, a simple statistical test is applied to determine whether the results obtained so far indicate a definite conclusion from the experiment, or whether more observations are needed to make the experiment decisive. The <$Isequential analysis;time savings realized by>experiment thus terminates as soon as a definite conclusion can be drawn, and the average number of observations required in experiments carried out in this manner tends to be definitely less than when the number has to be decided before hand. Consequently, this sequential method of performing comparative experiments has definite advantages, particularly when the observations are expensive or time consuming. This basic principle has been applied to Whole Body Counting where the "number of samples" taken is actually the number of seconds or minutes of <$Itime;counting>counting time, which is a <$Ivariable>variable and depends upon the bac<$Iunits;kilograms>kground, the amount of radioactivity in the person being counted, the level of radioactivity at which an alarm should be given, and upon the risks of failing to find true incorporation of radioactivity or of failing to state that a person is below this level of concern.<$Igraphs><$FThis paragraph and parts of several following paragraphs are quoted directly or paraphrased from "The Design and Analysis of Industrial Experiments," by Owen L. Davies, Chapter 3, pp. 57-98, Hafner Publishing Company, N.Y., 1956.>"

Most whole body counting work (and by <$Iextensions>extension, counting for external contamination or counting of wastes) is done for a pre-determined time. If a person has little or no internally deposited radioactivity due to his work, there really is no reason to spend any more <$Itime;counting>counting time than is necessary to prove that the level of radioactivity is definitely less than some pre-determined decision value. "An alternative and more economical procedure may be adopted when a series of short counts are made serially, that is to say, when the result of each separate trial is known before the next is carried out. In these tests the <$Itime;counting>counting time is not <$Imarkers>fixed in advance, but the test is applied to the accumulating data after each observation, the experiment being terminated as soon as a decision between the alternative hypotheses can be made with the desired degree of certainty."

data after each observation, the experiment being terminated as soon as a decision between the alternative hypotheses can be made with the desired degree of certainty."

"The intuitive sequential procedure discussed above can be put on an objective basis by a sequential significance test. The quantities required to plan such a test are precisely those needed for a non-sequential test. The test decides as each new observation comes to hand, in the light of all the information up to that time, whether:

@PARAGRAPHNUMB = 1.

@FOLLOWINGPARA = to accept the Null Hypothesis that no change of importance has occurred;<$&Zone1.TXT>

@PARAGRAPHNUMB = 2.

@FOLLOWINGPARA = to accept the Alternative Hypothesis that a real change has occurred; or<$&ZONE2.TXT>

@PARAGRAPHNUMB = 3.

@FOLLOWINGPARA = to continue taking observations.<$&Zone3.TXT>

The advantage of testing after each new observation, instead of reserving the decision until the completion of a <$Imarkers>fixed number, lies, as one might expect, in the smaller number of observations needed on the average to detect a given difference. Frequently this is only one-half of that required by the non-sequential test, and if an unexpectedly large effect occurs, the sequential test may yield a decision after only one or two observations."

@MINOR HEADING = The Sequential Test<$&FRSTGRAF.IMG>

"<$Isequential analysis;principle of>Sequential tests are best explained graphically. As each new observation is obtained, the value of the function of all observations recorded up to that time is calculated and plotted against the number of observations on a chart such as that shown in Figure <$R[C#,FRSTGRAF.IMG]3>-<$R[F#,FRSTGRAF.IMG]12>. On the chart are two boundary lines, the positions of which depend upon the risks a, b, of errors of the first and second kind <$Isequential analysis;Type 1 and Type 2 errors>(Type 1 and Type 2 errors), the magnitude of the difference it is important to detect, etc. The lines divide the chart into three zones:

@PARAGRAPHNUMB = 1.

@FOLLOWINGPARA = in which the Null Hypothesis is accepted, the lower right triangle of the graph;

@PARAGRAPHNUMB = 2.

@FOLLOWINGPARA = in which the Alternative Hypothesis is accepted, the upper left triangle of the graph; and

@PARAGRAPHNUMB = 3.

@FOLLOWINGPARA = in which there is no decision, the area between the two lines."

"The sequential test then consists in plotting the function of these observations f(x) against the number of observations n and continuing to take observations so long as the points fall within zone (3), the space between the lines. As soon as a point falls outside of this zone, that is, either in zone (1) (acceptance of the Null Hypothesis), or in zone (2) (acceptance of the Alternative Hypothesis), the observations are discontinued and the indicated decision is taken. The chosen function f(x) is used because of its powerful properties in discriminating between the hypotheses to be tested."

"In practice the chart can often be dispensed with, the boundary values being calculated in advance for each value of n. The test is then made by successive comparisons of the value of f(x) with the appropriate limits."

@MINOR HEADING = The <$Isequential analysis;single-sided alternative hypotheses>Single-Sided Alternative Hypotheses

In in vivo counting work or in the measurements of wastes we wish to test whether the population mean m of a series of observations is equal to some standard value m0 , which might be zero. We shall almost always be looking for a difference above the standard value m0. Thus, we shall be using the one sided limits which were discussed earlier in this chapter (see page <$R[C#,One-Sided Limits]3>-<$R[P#,One-Sided Limits]6>). Paraphrasing from Davies again, "Let d represent the difference it is important to detect, s the standard deviation [Editor's note: in this discussion the standard deviation is assumed to be known and constant], a the risk of asserting a significant difference when none exists, and b the risk of asserting no significant difference then the mean value is really m1 = m0 + d ."

"The function f (x) plotted for this test is simply the total T of the observations up to the time considered, and the boundaries are parallel straight lines with slope s, cutting the axis of T at h1 and h0. The values of h0, h1, and s are given by the equations

@EQUATION = <$Eh sub 0~=~{-b~sigma sup 2} over delta>, <$Eh sub 1~=~{a~sigma sup 2} over delta>, <$Es~=~mu sub 0~+~delta over 2>

where

@EQUATION = <$Ea~=~roman {ln}^left (~{~(1~-~ beta~)} over alpha~right )> and <$Eb~=~roman {ln}^left (~{~(1~-~ alpha~)} over beta~ right )> ."

@BODYTEXTPGBRK =

@MINOR HEADING = An Example<$&ORIGDATA.TXT>

<$Isequential analysis;an example>Let us use data from Davies, (op.cit.). Let us quote from his paragraph 3.211, page 60:

@BODYTEXTPGBRK = "It is instructive to compare the sequential and non-sequential types of test upon the same data. An inspection scheme for primers for explosives was required. It was assumed that s was constant and equal to 0.03, and the scheme was planned so that there would only be a small risk (a = 0.01) of accepting a bad batch, that is a batch with mean density as low as m0 = 1.50 g/cc, and a small risk (b = 0.02) of rejecting a good batch, that is a batch with mean density m1 = 1.54 g/cc. Thus d = 0.04, s = 0.03, a = 0.01, and b = 0.02. It was found that for the non-sequential test eleven observations would be required and that the test should be made by calculating the mean of the sample of eleven (Davies, paragraph 2.43), rejecting the batch if the mean was less than 1.532 g/cc and accepting it otherwise." These data are shown in Table <$R[C#,ORIGDATA.TXT]3>-<$R[T#,ORIGDATA.TXT]5>, below. The top section of the table shows the conditions as defined in this paragraph. The middle section, "Uncoded Data," shows the original data in the first two columns. We shall discuss this first.

Table <$R[C#,ORIGDATA.TXT]3>-<$R[T#,ORIGDATA.TXT]5> shows the calculations of T0, T1, and the sum of the individual measurements. Figure <$R[C#,DAVIESUN.IMG]3>-<$R[F#,DAVIESUN.IMG]13>, below, shows how these data look in a graph. "It will be found in practice that if m0 is large compared with d these boundary lines will rise very steeply and appear to be very close together, so the chart will be difficult to use (Davies, op. cit., p 61)." Looking at the figure we can see that Davies was right. The chart is of not much use. We can, however, compare the data in the "Cumulative" column with the T0 and T1 values. We see that the fifth sample caused the cumulative value to be greater than the T1 value, thus ending the sequential tests."

"Since the test is to detect a difference, it will not be affected if a constant amount is subtracted from each observation. For purposes of convenience, therefore, instead of considering the actual density we consider the amount by which the density exceeds 1.40, that is to say, 1.40 is subtracted from each observation.

@BODYTEXTPGBRK = Then m0 = 0.10, m1 = 0.14, h0 = -0.087784, h1 = 0.10316, and s = 0.12. See the top part of Table <$R[C#,ORIGDATA.TXT]3>-<$R[T#,ORIGDATA.TXT]5> where we have listed these values in the column labeled "Coded." The graph of these coded data is found on the next page, Figure <$R[C#,DAVIESCD.IMG]3>-<$R[F#,DAVIESCD.IMG]14>. <$&DAVIESUN.IMG>

Now we can see how the cumulative sums of the coded values proceed to make the decision to "Accept the Batch" after the fifth sample. Remember, that when Davies performed a non-sequential test of the same data, eleven samples were required to make the same determination.<$IESC key><$&DAVIESCD.IMG>

The reader can readily observe that if d, the difference it is desired to detect, becomes smaller, the values of h0 and h1 will become farther apart and it will take longer to make a decision. Also, smaller Type 2 errors, b, will push the line through h0 farther from the center (the "zero" value) and will make us less ready to reject the batch. Similarly, a small value for the Type 1 error, a, will make h1 larger and make us less ready to accept the batch.

"Those persons newly introduced to sequential tests are often worried by the thought that a sequential test might never terminate, or at least that it might go on for very much longer than the experimenter was willing to contemplate. In practice this is not a serious difficulty, for it can be shown that the probability that a sequential test would require twice or three times as many observations as the corresponding non-sequential test is exceedingly small; usually, the number of observations required for the sequential test is less than for the non-sequential test. Note the non-sequential and sequential methods are used for deciding which is the more proficient of two opponents in games of skill. The scoring system in "darts," for example, is non-sequential, while in "tennis" at a certain stage (when the score 40-all or deuce is reached) a sequential scoring system is used, it being necessary for one of the two players to lead by two points to win. Just as in practice it is almost certain that a game of tennis will end after a few exchanges, so in sequential significance tests there is no serious risk of the experiment being unduly prolonged."

@BODYTEXTPGBRK = The Helgeson software has a parameter which controls the length of the total count. If a decision cannot be made within that time period, the computer states "No decision, try re-counting," and the results are calculated based on the information at hand.

@MINOR HEADING = Sequential Analysis and the <$IPoisson Distribution;demonstration program>Poisson Distribution<$&SEQNTL01.TXT>

In the previous example s was assumed to be known and constant. In radioactive counting work s may be known, but, since we shall be taking <$Isequential analysis;demonstration program>serial counts and adding them together, it will not be constant. Therefore, we must use the general ideas of sequential analysis but recognize that sigma, s, will change as we progress further in the counting series. This means that our values h0 and h1 and T0 and T1 will also change. The balance of the testing is identical to the example presented earlier. Table <$R[C#,SEQNTL01.TXT]3>-<$R[T#,SEQNTL01.TXT]6>, below shows data produced by a p<$Iproportional counters>roportional counter operated in the Sequential Analysis mode.<$&SEQNTL10.IMG> The initial conditions are given in the first 11 lines and then data from three runs are given. Figure <$R[C#,SEQNTL10.IMG]3>-<$R[F#,SEQNTL10.IMG]15>, above, is a graphical presentation of the first group of these data. Note that the process took nine seconds before the results showed that the count was below the <$Iaction point>action point of 6 cps.<$&SEQNTL11.IMG> Figure <$R[C#,SEQNTL11.IMG]3>-<$R[F#,SEQNTL11.IMG]16>, below, shows a graphical presentation of <$Isequential analysis;demonstration graphs>the second group of these data. Note that this time the process took eleven seconds before the results showed that the count was below the <$Iaction point>action point of 6 cps.

Figure <$R[C#,SEQNTL12.IMG]3>-<$R[F#,SEQNTL12.IMG]17>, above, <$&SEQNTL12.IMG>is a graphical presentation of the third group of these data. Note that now the process took only five seconds before the results showed that the count was below the <$Iaction point>action point of 6 cps. These are the types of results you should expect from sequential analysis. Not every count will require the same amount of time, but you can be sure that the counts will meet the defined requirements of <$Iaction point>action point, Type 1 and Type 2 error because the laws of mathematics cannot be broken.

The same types of <$Igraphs>graphs will be obtained if the <$Iaction point>action point was routinely exceeded: not all of the counts would take the same amount of time, but if the subject was truly above the <$Iaction point>action point, the results would so indicate conforming with the Type 2 error.

As we stated on page <$R[C#,DAVIESCD.IMG]3>-<$R[P#,DAVIESCD.IMG]30>, if a decision cannot be made within the pre-defined maximum <$Itime;counting>counting time period, the computer states "No decision, try re-counting," and the results are calculated based on the information at hand. Thus, the System Manager may determine the maximum length of time for a count. Obviously, when the <$Iaction point>action point is closer to the bac<$Iunits;kilograms>kground, a longer count time is required in order to arrive at a decision as has already been drawn.

The resulting time savings is very significant and the reduction in manpower requirements during a major outage are impressive. The obvious advantage of this type of a counter lies in the fact that one may obtain adequate counting information in a very short period of time. Thus, the requirements of regulatory authorities are being met while at the same time a considerable amount of time away from productive work is saved.

One will recognize that the statistical principles discussed above apply to any type of whole body counting instrument. This is not unique to the particular counter built by the author's company. The same fundamental rules of statistics will apply to all of these counters, namely, that the longer the <$Itime;counting>counting time of the subject and/or bac<$Iunits;kilograms>kground, the lower will be the <$Isensitivity;analytical>sensitivity of the counter.