Home About us Editorial board Search Ahead of print Current issue Archives Submit article Instructions Subscribe Contacts Login 
Home Print this page Email this page Small font size Default font size Increase font size Users Online: 287

 Table of Contents 
Year : 2019  |  Volume : 42  |  Issue : 1  |  Page : 15-21  

A new look at blind test procedures in personnel monitoring

1 Radiological Physics and Advisory Division, Bhabha Atomic Research Centre, Mumbai, Maharashtra, India
2 Department of Physical Sciences, Homi Bhabha National Institute, Mumbai, Maharashtra, India

Date of Submission19-Feb-2019
Date of Decision16-Mar-2019
Date of Acceptance25-Mar-2019
Date of Web Publication3-Jun-2019

Correspondence Address:
Sneha Chandrasekhar
Radiological Physics and Advisory Division, Bhabha Atomic Research Centre, Anushaktinagar, Mumbai - 400 094, Maharashtra
Login to access the Email id

Source of Support: None, Conflict of Interest: None

DOI: 10.4103/rpe.RPE_7_19

Rights and Permissions

In India, a single type of dosimeter is used all over the country for estimation of occupational doses due to X, beta, and gamma radiations. The quality assurance program of external personnel monitoring in India includes an annual announced test and periodic blind tests. The announced test carried out by a central agency includes various energies and angles of irradiation. Blind tests are carried out by specified users of personnel monitoring service who are distributed throughout the country. Due to the lack of easy availability of various energy sources, blind tests are mostly carried out for a single energy. Analysis of the results is carried out by means of ANSI criteria or trumpet curve analysis which may not be sufficiently stringent given that a single energy and angle are used. In view of this and also to meet the specific requirements of the Indian monitoring scenario, keeping in mind existing practices, an attempt is made to have a new look at the analysis of blind test results. The criteria which have been considered are zeroing of dosimeters, coefficient of variation, and bias. The results of a blind test based on harmonized procedures can be utilized to understand the actual quality of service and also identify deviations in procedures, if any.

Keywords: Blind test, coefficient of variation, personnel monitoring

How to cite this article:
Chandrasekhar S, Pradhan S M, Bhattacharya M, Bakshi A K, Datta D. A new look at blind test procedures in personnel monitoring. Radiat Prot Environ 2019;42:15-21

How to cite this URL:
Chandrasekhar S, Pradhan S M, Bhattacharya M, Bakshi A K, Datta D. A new look at blind test procedures in personnel monitoring. Radiat Prot Environ [serial online] 2019 [cited 2020 Jul 13];42:15-21. Available from: http://www.rpe.org.in/text.asp?2019/42/1/15/259673

  Introduction Top

In India, a single Thermo luminescence (TL)-based dosimeter developed by Bhabha Atomic Research Centre (BARC) is approved for personnel monitoring of all radiation workers.[1] The performance characteristics of this dosimeter are well established.[2],[3] There are several accredited monitoring laboratories for providing monitoring service using this dosimeter. The quality of service provided is assured by the central accrediting agency by ensuring that the monitoring laboratory is knowledgeable about the correct procedures and is capable of providing service as per the accreditation requirements. The quality is further confirmed by a regular announced comprehensive quality assurance (QA) program. In addition, blind tests are also periodically carried out in the country for several of the accredited monitoring laboratories by the users of monitoring services, mainly health physics personnel.

During blind tests, the monitoring laboratory is unaware about the test being carried out. Therefore, the test has the merit of being a true representation of the actual service. However, since such a test is not carried out by a central agency, several different agencies are involved in carrying out the tests. The irradiations for the blind tests are made in laboratory conditions and do not include either angles or irradiations at energies other than137 Cs gamma.

Internationally, documents compiled for assessing the performance of monitoring laboratories attempt to simulate field like situations by taking into account variables such as energy dependence, fading, and angular response. At present, trumpet curve analysis[4] and ANSI[5] criteria are used for determining the satisfactory performance of a laboratory.[6] These criteria for assessment also take into account all the variables mentioned above. In India, since a single type of dosimeter is used, none of these variables exist or the variations are identical, i.e. identical results are expected for all processors. Where blind tests are concerned, it is possible that these methods are not stringent enough since irradiations do not encompass all energies and angles.

Therefore, keeping in mind existing practices of using a single photon energy and single angle of irradiation, an attempt is made to understand the reasons for variations in bias and coefficient of variation (CoV). A test procedure is proposed which is similar to the existing procedures but fixes variables such as number of dosimeters exposed to each dose level.

  Proposed Procedure for Blind Testing Top

For functional convenience, blind tests are usually carried out by the agencies using the personnel monitoring services, generally health physics personnel. Due to a lack of availability of different calibrated sources of radiation, irradiations in general are carried out only with137 Cs energy and at normal incidence.

Since angles and energies are not being varied, the proposed testing is for the variables which can be tested, that is, the quality of dosimeters used, the performance of the instruments, and the method adopted for calibration. These can be checked by testing for zeroing of dosimeters, CoV among dosimeters, and the accuracy of calibration.

  Zeroing of Dosimeters Top

Before assignment and issue of dosimeters, they are annealed in an air-circulated oven at 230°C for 4 h for zeroing. The annealing treatment removes residual TL of previous irradiation and helps in establishing the sensitivity of the dosimeters. After annealing, the laboratories are instructed to read a sample of dosimeters[7] (about 5%) to confirm the adequacy of annealing. External annealing is required since the readout in the reader removes typically only about 90% of the TL. When the reading of a single disc is very high (>50 mSv), the external annealing is inadequate to completely reduce the background reading to below acceptable values. In such cases, all laboratories are advised to remove such dosimeters from service.

The issues that exist with reference to annealing are (i) a failure of oven, (ii) failure of procedure due to which some cards are not annealed, and (iii) nonremoval of dosimeters which may have a high background reading. Therefore, a test for adequacy of annealing may be necessary as a part of blind test.

The background dose for the period of dosimeter use is arrived at based on the readings of the control dosimeters. Monitoring frequency at power plants is monthly, and about 10 dosimeters are used as control, the average reading of which is taken as the value of background dose for subtraction. Dose estimation from dosimeters assigned to individuals is carried out after subtraction of this control reading from the individual dosimeter readings. The minimum detection limit at 95% confidence is 0.1 mSv.[8] Therefore, theoretically, a false positive, i.e., a dosimeter registering dose 0.1 mSv or greater when it has not been given the dose is possible 5% of times. However, since most of the dosimeters during service are exposed to doses of the level of the background radiation, readout of the dosimeter in the reader is nearly equivalent to annealing in the oven as far as the readings of the Thermoluminescence Dosimeter (TLD) card are concerned. That is, if dosimeters have been exposed to only a single month's background radiation (≈50 μSv), then a second readout of this card may not be distinguishable from the reading of a card which has been annealed in oven [Table 1]. Therefore, it may not always be easy to detect the absence of annealing from a random sample of cards.
Table 1: Typical total readings of dosimeters (μSv) read in a calibrated reader

Click here to view

However, if a large number of dosimeters have been exposed to significant doses in previous monitoring period, the probability of detection becomes somewhat higher. Quantification in such situations is very difficult and not likely to be accurate.

It should also be recognized that an error such as a failure in annealing of used dosimeters will not be an isolated incident affecting only a few dosimeters but would mean a disruption in the flow of work in the monitoring laboratory, which may become evident to the monitoring laboratory even before the reading of service dosimeters and remedial action can be taken accordingly. It should be remembered that the monitoring laboratory has the facility of viewing the glow curve which may help in identification of such an occurrence easily – a facility which is not available to the external agency.

A failure of annealing, therefore, may be very difficult to detect by an external agency, without the knowledge of the monitoring laboratory, with a small number of dosimeters. However, it may still be a good idea to check the performance of the laboratory at “0” dose level, for checking the quality of dosimeters as far as “zeroing” is concerned.

  Coefficient of Variation Top

Estimation of coefficient of variation for small samples

The standard deviation (SD) and the CoV are important parameters for testing an individual monitoring system. The dosimeters in a monitoring laboratory at the time of introduction are expected to have a CoV <0.05 at dose level of 3 mSv. Groups of dosimeters are created at the time of procurement which differ in sensitivity by <5%, and individual calibration factors for the dosimeters are not used. With use, it is expected that the CoV increases from its original value. It is expected that the monitoring laboratory discontinues use of dosimeter batches whose CoV degrades to above 0.05. However, conditions such as delay in identification of deterioration in quality of dosimeters or delay in supply or procurement of fresh batch of dosimeter may occur wherein the monitoring laboratory is continuing service with dosimeters with CoV >0.05. In this study, a test value of 0.10 is used for checking the CoV of dosimeters in service. Initial use of this high value may help in arriving at some conclusions regarding performance of various monitoring laboratories before scaling down of this value further to internationally acceptable values of CoV.

In a blind test, since only a small sample is used, the measured SD (s) can have widely varying values. Furthermore, the most probable value of s is less than the conventional true value (σ).[9] If 10 dosimeters are used in a single sample, the most probable value of s is 0.943 σ. Therefore, as per International Electrotechnical Commission (IEC)[10] requirements, testing for CoV is based on a probability of 50% of passing, i.e., passing will indicate a 50% probability that the observed CoV is less than or equal to the required value of CoV.

However, in the present context where a CoV of 0.10 is used as test value, it may be more relevant to have a probability of failing of 95% if the CoV is equal to 0.10. By the same methods as used in Brunzendorf and Behrens,[9] the values as given in [Table 2] are obtained for a single test for different sample sizes.
Table 2: Values of coefficient of variation for various sample sizes

Click here to view

It can be seen from [Table 2], that for smaller sample sizes, the probability of failing is larger even when the sample has true value of CoV less than the acceptable. Therefore, smaller samples are prone to greater errors. A larger sample size should always be preferred. Further, for a sample size of 5, if the true value of CoV is 0.05, the probability of failing is more than 50% making this more stringent than required. Therefore, a single test of this nature may be too stringent and unacceptable. However, since blind tests are carried out on a periodic basis (frequently monthly), the results of the testing for each period can be considered to be a sample, with a maximum of 12 samples being obtained for each year. The results of each test should be viewed in relation to the entire year's result. IEC[10] gives c1 and c2 values as multipliers for the CoV based on sample size and number of tests. Again, since these values are derived for a 50% probability of passing, they cannot be used directly. Therefore, the values of “c,” factor for multiplication with CoV for multiple tests, are calculated using the same methods as described in Brunzendorf and Behrens[9] for a 5% probability of passing if the relative SD is 0.10.

If tests are performed with the criteria for passing as 0.10 multiplied by the relevant factor “c” mentioned in [Table 3], the probability of passing the test for a batch of cards with CoV of 0.10 is 5%. The probability of passing the test for a batch of cards with CoV 0.08 and 0.05 also vary as shown in [Figure 1]. It is evident that multiple tests improve the possibility of passing when the CoV is below 0.10. Furthermore, best, unambiguous results are obtained for larger samples whereas the probability of passing for a batch of cards with CoV of 0.10 continues to be 5% irrespective of number of tests and sample size.
Table 3: Acceptable values of “c” for multiple tests of standard deviation for a 5% probability of passing if coefficient of variation=0.10

Click here to view
Figure 1: Probability of passing for various CoV. (a) For sample size 5, (b) For sample size 10, and (c) For sample size 20. Filled squares are for samples with CoV of 0.05, filled circles for samples with CoV of 0.08, filled triangles for samples with CoV of 0.10, and crosses for samples with CoV of 0.12. The X-axis indicates the number of samples and the Y-axis gives the probability of passing. CoV: Coefficient of variation

Click here to view

Level of dose for estimation of coefficient of variation

As per IEC,[10] while testing for the CoV, the doses to which the dosimeters are to be exposed range from 0.1 to 1000 mSv. The CoV acceptable for each dose varies with the level of dose, higher values being permissible for lower doses. The exact variation of CoV with given dose “H” is given by the following equation:

However, for doses >1.1 mSv, the maximum acceptable value is 5% and remains the same up to the highest dose in the measurable range. The SD of the Indian TLD system has been found to vary as per the following equation:[11]

Where σT is the total SD, σB is the SD of background dosimeters, σμ is the relative SD at high doses, and KT is the value of dose equivalent.

A comparison of the SDs as given by equation 2 (using σB= 15 μSv and σμ= 0.05) and SDs derived from the IEC values of maximum acceptable CoV is given in [Figure 2]. The SD has been calculated from the CoV using the following equation:
Figure 2: Plot of variation of standard deviation with dose. Squares indicate standard deviation obtained from IEC requirement of CoV. Crosses indicate standard deviation for TLD system as per Equation 2. CoV: Coefficient of variation

Click here to view

Such a comparison may not be rigorously correct since the purpose of defining limits to CoV by IEC is different; however, since such a comparison does shed some light on the behavior of SD, the same is being done. Equation 2 predicts that the SD remains nearly the same at very low doses as at “0” dose due to the predominance of the first term with a dose-dependent increase at higher doses. However, in case of IEC, extremely low values of SD are required at very low doses, with higher values being acceptable as dose increases. It can be seen from [Figure 2] that between doses equivalents of 0.40 mSv and 1.1 mSv, the increase in SD with dose is very minor for the IEC curve, with some slight decrease in-between, and hence, it is observed that between 0.15 mSv and 1.0 mSv, the difference between IEC requirements and observed values is very high. An implication of such an observation is that when the given dose is <1.0 mSv, dosimeters with CoV at high doses of >5% may also show acceptable values of CoV at these dose values.

In view of this, for blind testing, it may be required to define values of CoV for doses <1.1 mSv based on equation 2. These values can be calculated as shown in [Table 4] with suitable values of σμ and σB. In the particular case discussed of blind testing, typical values could be σμ= 0.10 and σB= 0.015 mSv. Since the detection limit of the system is 0.1 mSv, the values of CoV at lower doses do not have any significance.
Table 4: Values of coefficient of variation at doses below 1.1 mSv

Click here to view

As can be seen from [Table 4], the values of CoV do not significantly deviate from 10% even for lower doses. Therefore, by avoiding doses <400 μSv, a single test value of 10% may be used for all doses.

  Bias Top

A single point calibration of the system is carried out before the reading of service dosimeters to ensure continued reliable performance of the card reader system. Calibration factors in the reader are set accordingly. The bias will include the following:

  1. Ten dosimeters are used for calibration. This will be a random sample from the batch of cards used in service. If the dosimeters have a CoV of 10% at delivered dose of 3 mSv, a random sample of 10 dosimeters will have dispersion in their means of 3.16% (s1). At 95% confidence, the half width of the confidence interval of the mean will be given by U1= t9× s1 where t9 is the student's t factor for a sample of size 10 at required confidence
  2. Dosimeters used in the blind test will be another random sample from the same population. The dispersion in the means of these samples will again be 3.16% (s2) if 10 dosimeters are used in the blind test, and the given dose is equal to or >1.1 mSv. At 95% confidence, the half width of the confidence interval of the mean will be given by U2= t9× s2 where t9 is the student's t factor for a sample of size 10 at required confidence
  3. The control value is obtained from the reading of 5 dosimeters for calibration, a total of 15 readings, and at least 10 dosimeters for reading of blind test cards. This includes corrections for both natural background dose and zero dose indication. The SD of the control background dosimeters is generally between 10 and 20 μSv. In case of calibration, assuming a value of 15 μSv and at 95% confidence, the half width of the confidence interval of the mean will be given by U3= t14× sb where t14 is the student's t factor for a sample of size 15 at the required confidence. This gives a negligibly small value as compared to U1 and U2. Therefore, the error due to control subtraction is neglected in all further calculations
  4. A further source of error is the rounding off of doses routinely practiced while reporting. All laboratories report the evaluated dose after rounding off to the nearest 0.05 mSv. If a rectangular distribution is assumed, this gives an error U4 of 14 μSv irrespective of dose for each dosimeter.

Adding the errors in quadratures, and assuming that there is no error in the delivery of the dose, the bias at 95% confidence for various sample sizes and doses can be calculated and is given in [Table 5]. As discussed previously, the CoV is higher for lower doses. However, since a high test value of CoV of 10% is used, the same is used in calculations for all dose levels.
Table 5: Maximum acceptable bias for different sample sizes

Click here to view

  Summary of Blind Test Top

In summary, a blind test could follow these steps:

  1. The results of blind testing carried out each month or at any other periodicity will be treated as a single sample
  2. The easiest approach is to give the same dose for all dosimeters in a single sample with variation of dose between samples (months/periodicities)
  3. The bias will be calculated as true dose-average reported dose for all blind test dosimeters. The bias for any sample should not exceed the relevant value given in [Table 5]
  4. The reference value of CoV multiplied with the relevant factor “c” taken from [Table 3] will be the maximum acceptable value of CoV
  5. In no test should the measured value of CoV exceed the value obtained in step no 4
  6. Testing for “zero dose” though difficult to detect should nevertheless be carried out. Larger numbers of dosimeters will give greater probability of detection. Detection of even a single occurrence irrespective of number of dosimeters used for testing should be viewed seriously.

  Discussion Top

In view of the above discussions and keeping in mind the specific requirements of the Indian system, alternative methods for carrying out the blind testing can be used.

A uniform approach to the procedure of blind testing will aid in arriving at conclusions regarding the performance of any monitoring laboratory. Though the values of CoV used are somewhat higher than what is internationally acceptable, these can be used till some conclusion can be reached about the actual quality of service. The factors arrived at in the study can easily be scaled down or modified, if required.

ANSI 1993[12] states – “Testing of newer programs may not benefit from blind tests since steps are under routine computer control and statistical evidence of performance on routine quality control tests may be equally valid”. This implies that regular laboratory generated values of SD are equally useful for arriving at values of CoV at which the individual laboratory is functioning. However, test results arrived at externally have a greater impact on a lay audience and blind tests therefore continue to have a large patronage.

  Conclusions Top

The external QA program is a very important tool to ensure the continued reliable performance of various monitoring laboratories. Keeping in mind the specific requirements of the Indian monitoring system, a highly specific protocol has been generated. This protocol should satisfy the primary requirements of the Indian personnel monitoring program and also be in line with the international requirements.

In view of the fact that blind tests are being conducted regularly, the proposed method of testing will give additional information regarding the performance of individual laboratories. The agencies carrying out the blind tests being varied with limited access to different photon energy sources, irradiation of dosimeters to a single energy and angle is the easiest approach to testing. This approach to the routinely carried out program of blind testing can be of use to identify lacunae either in the performance or in the knowledge of individual laboratories and help in improving performance.


Authors are thankful to Dr. Pradeepkumar K S., Director, Health, Safety and Environment Group, BARC for his encouragement in the work.

Financial support and sponsorship


Conflicts of interest

There are no conflicts of interest.

  References Top

Vohra KG, Bhatt RC, Chandra B, Pradhan AS, Lakshmanan AR, Shastry SS, et al. A personal dosimeter TLD badge based on caSO4:Dy teflon TLD discs. Health Phys 1980;38:193-7.  Back to cited text no. 1
Pradhan SM, Sneha C, Chourasiya G, Adtani MM, Tripathi SM, Singh SK. Development of an algorithm for evaluating personal doses due to photon fields in terms of operational quantities for TLD badge system in India. Radiat Prot Dosimetry 2009;136:176-84.  Back to cited text no. 2
Bakshi AK, Srivastava K, Varadharajan G, Pradhan AS, Kher RK. Development of an algorithm for TLD badge system for dosimetry in the field of X and gamma radiation in terms of Hp(10). Radiat Prot Dosimetry 2007;123:148-55.  Back to cited text no. 3
ISO-14146 Radiation Protection — Criteria and Performance Limits for the Periodic Evaluation of Processors of Personal Dosemeters for X and Gamma Radiation. ISO 14146; 2000.  Back to cited text no. 4
American National Standards Institute. American National Standard for Dosimetry – Personnel Dosimetry Performance Criteria for Testing. American National Standards Institute; 2009.  Back to cited text no. 5
Singh VP, Managanvi SS, Bihari RR, Bhat HR. Operational experience of electronic active personal dosemeter and comparison with CaSo4:Dy TL dosemeter in Indian PHWR. Radiat Prot Dosimetry 2013;156:93-102.  Back to cited text no. 6
Datta D, Palani Selvam T, Bakshi AK, Pradhan SM, Sneha C, Srivastava K. Handbook on TLD-Based Personnel Monitoring (REV. 1-2018). BARC/2018/E/007: Radiological Physics and Advisory Division; 2018.  Back to cited text no. 7
Sneha C, Pradhan SM, Adtani MM. Study of minimum detection limit of TLD personnel monitoring system in India. Radiat Prot Dosimetry 2010;141:168-72.  Back to cited text no. 8
Brunzendorf J, Behrens R. How to type test the coefficient of variation of an indication. Radiat Prot Dosimetry 2007;123:21-31.  Back to cited text no. 9
International Electrotechnical Commission IEC 62387. Radiation Protection Instrumentation – Passive Integrating Dosimetry Systems for Personal and Environmental Monitoring of Photon and Beta Radiation. 1st ed. International Electrotechnical Commission; 2012.  Back to cited text no. 10
Sneha C, Pradhan SM, Madhumita B, Pradeep R, Datta D. Study on coefficient of variation in personnel monitoring system in India. Nucl Technol Radiat Prot 2016;31:388-92.  Back to cited text no. 11
American National Standards Institute. American National Standard for Dosimetry – Personnel Dosimetry Performance Criteria for Testing. American National Standards Institute; 1993.  Back to cited text no. 12


  [Figure 1], [Figure 2]

  [Table 1], [Table 2], [Table 3], [Table 4], [Table 5]


Similar in PUBMED
   Search Pubmed for
   Search in Google Scholar for
 Related articles
Access Statistics
Email Alert *
Add to My List *
* Registration required (free)

  In this article
Proposed Procedu...
Zeroing of Dosim...
Coefficient of V...
Summary of Blind...
Article Figures
Article Tables

 Article Access Statistics
    PDF Downloaded161    
    Comments [Add]    

Recommend this journal