EPR dosimetry intercomparison using smart phone touch screen glass

Abstract

This paper presents the results of an interlaboratory comparison of retrospective dosimetry using the electron paramagnetic resonance method. The test material used in this exercise was glass coming from the touch screens of smart phones that might be used as fortuitous dosimeters in a large-scale radiological incident. There were 13 participants to whom samples were dispatched, and 11 laboratories reported results. The participants received five calibration samples (0, 0.8, 2, 4, and 10 Gy) and four blindly irradiated samples (0, 0.9, 1.3, and 3.3 Gy). Participants were divided into two groups: for group A (formed by three participants), samples came from a homogeneous batch of glass and were stored in similar setting; for group B (formed by eight participants), samples came from different smart phones and stored in different settings of light and temperature. The calibration curves determined by the participants of group A had a small error and a critical level in the 0.37–0.40-Gy dose range, whereas the curves determined by the participants of group B were more scattered and led to a critical level in the 1.3–3.2-Gy dose range for six participants out of eight. Group A were able to assess the dose within 20 % for the lowest doses (<1.5 Gy) and within 5 % for the highest doses. For group B, only the highest blind dose could be evaluated in a reliable way because of the high critical values involved. The results from group A are encouraging, whereas the results from group B suggest that the influence of environmental conditions and the intervariability of samples coming from different smart phones need to be further investigated. An alongside conclusion is that the protocol was easily transferred to participants making a network of laboratories in case of a mass casualty event potentially feasible.

Appendices

Appendix 1

A data point that is graphically far from the relationship between x and y described by the other points generally deserves to be further investigated. In the case of the present exercise, the 0.8-Gy calibration data point was far from the linear pattern of the other data points for most participants (see Fig. 1, lower panel) and required further investigation. For this reason, although keeping in mind the weak statistical power of the sample, specific statistic tests were carried out to identify outliers, leverage points, and influential points of the calibration curves (Draper and Smith 1998; Chatterjee and Hadi 1986):

• a. Outliers are data points that appear to deviate markedly from other members of the sample in which it occurs, as defined by Grubbs (1979). In this exercise, outliers were identified by the Grubbs test (Grubbs 1979) and by the deleted studentized residuals (Draper and Smith 1998). One individual measurement (out of the three replicates of participants 3, 6, and 9) for the dose 0.8 Gy exceeded slightly the Grubbs test value (assumed as 2).

• b. Leverage points are the points of the independent variable (in the present analysis, the calibration dose) having the potential to dominate a regression analysis, but not necessarily to influence it. These were identified according to the Sokal test. The 0.8-Gy calibration dose points resulted not to cause large changes in the linear fit parameter estimates when they were deleted. The only dose point with a statistically high leverage value was the 10-Gy dose point.

• c. Influential points are data points that greatly affect the parameters of the regression line and therefore deserve further investigation. They are typically outliers weighted for their leverage value. The influential points of the calibration curves were identified by the Cook distance test (Cook and Dennis 1979). This parameter was evaluated for the calibration curve of every participant, and it was repeated after removing one calibration data point at a time point from all the calibration curves. In Fig. 3 (top), the resulting estimated slopes versus estimated intercepts obtained by removing one calibration data point at a time is plotted. The estimated coefficients are all bunched together regardless of the removed data points, except for the calibration curve of participant 5 when the points 0.8, 4, and 10 Gy were eliminated, and for the curves of participants 5, 6, and 10 when the 0-Gy calibration point was eliminated. The Cook distance (Fig. 3, bottom) indicated that only the 10-Gy point was influential for some participant (2, 7, and 9). This was an expected result given the fact that the 10-Gy calibration dose is an extreme point of the calibration curve and obviously a high leverage point. Although this test is controversial and there are different opinions on the threshold to be chosen for the Cook distance, it is possible to state that no influential points were found.

  Fig. 3

  

  Top estimated slopes versus estimated Y-intercepts obtained by removing one calibration data point at a time. Bottom Cook’s distance test

Although no outliers were identified, it is out of doubt that a problem existed in some participants’ measurements in the signal line shape at 0.8 Gy, which appeared different from the signal observed by the other participants. Figure 4 shows the comparison of the spectrum for 0.8 and 10 Gy for participants 5 and 13. Whereas the signal at 10 Gy appeared similar in shape and intensity (although within an expected intersample variability), the signal for the 0.8-Gy irradiated sample was evidently different in line shape and intensity. Excluding the 0.8-Gy calibration data point from the calibration curve of some participants, the statistics indeed improved significantly. For instance, for participant 5, the CL and DL would drop to 2.3 and 4.6 Gy, respectively, and Pearson’s r would increase to 0.953. Participant 10 also showed a signal line shape different from others at all doses, and this is the reason why it was not possible to identify any influential point for this participant. Reasons for this will have to be further investigated.

Fig. 4

EPR spectra of the samples irradiated at 0.8 Gy (top) and 10 Gy (bottom) for participants 13 (left) and 5 (right)

Appendix 2

After distributing the samples to participants of group B, the remaining part of the samples was stored in the laboratory of participant 5, in a cabinet, i.e., as far as possible in the absence of light. This turned out to be a lucky fortuity. When it became clear that the participants of group B were measuring large fluctuations in the calibration samples, those remaining fragments were measured by participant 5. The measurements were carried out 30 days after irradiation and repeated after further 30 days. The calibration curves are shown in Fig. 5. The fit error in the calibration curve appeared to be significantly smaller than that of Table 1. The estimated values for the blind doses were as follows: 0.69 ± 0.375, 1.16 ± 0.62, 3.46 ± 0.35, −1.2 ± 0.358, for the 0.9, 1.3, 3.3, and 0 Gy doses, respectively, showing satisfactory agreement between the actual and the measured doses for the blind test. Although these data should be taken with prudence because they were measured one and 2 months after irradiation, they are an indication that inexperience may be only partly responsible for the bad performance of participants of group B. Albeit very well experienced, participant 5 performed differently between the intercomparison and when using samples which had not been exposed to light.

Fig. 5

Calibration curve obtained by participant 5 with samples of glass taken from the same batch used for group B, but stored under controlled conditions and in condition of absence of light (as far as possible)