References : [1] Hurkmans CW, et al Set-up verification using portal imaging; review of current clinical practice. Radiother. Oncol. 2001;58: [2] Greer PB, et al. Comparison of two methods for anterior-posterior isocenter localization in pelvic radiotherapy using electronic portal imaging. Int. J. Radiat. Oncol. Biol. Phys. 1998;41: Is it valid ? It seems too easy. Such a method does not take account of out of plane-rotations, non-rigid movements, observer variability, etc. In fact the method does take into account all of these variations. Any factor which may result in a measurement yielding different results is included in the  U for the measurement method. Unlike reproducibility studies this method incorporates all possible sources of measurement error. The method is not perfect however. The following assumptions and limitations apply. 1) That each measurement method is truly independent. 2) That each measurement method is related to the same definition of the patient position. 3) That both patient position variation and measurement uncertainty can be described by gaussian distributions. The first of these means that the method could not be used to directly compare two methods which both rely heavily on the same underlying parameter (such as digital couch scale vs. manual couch scale). The second means that one could not for example compare a method which relies on the position of the pelvic bones with a method which detects implanted seeds, since the genuine position (P) is not the same for each case. In this analysis all methods relate ultimately to the position of the pelvic bony frame. Conclusions : Measurement imprecision can form a significant component of measured isocentre reproducibility data. When comparing results from different centres, or even machines in the same centre it may be that the tools or training provided for the analysis of the images have as much to do with the final result as the actual patient set up variation. In this example the analysis gave strong evidence that the A/P set up results were not optimal and could be improved by moving to daily set up using the digital couch scale, confirming the result of Greer et al [2]. This also showed that the portal analysis was not sufficiently accurate to improve patient set up using daily on-line correction. This method is a useful tool in determining the quality not just of the set up but also of the portal image analysis. Furthermore it is a reminder not to take the result of the portal image analysis as truth, but another measurement with its own inherent uncertainty. Methods : Portal images were acquired at intervals throughout treatment for 112 patients undergoing prostate radiotherapy from November 1997 to July Patients were positioned from tattoos. At each imaging fraction further independent measurements were also made to indicate the patient position. In the sup-inf direction the position was determined from the lateral and anterior portals. In the A/P direction the position was determined from the lateral portal, the anterior SSD, the digital couch scale and a manual couch height reading with a ruler. Because the two couch scale measures are not truly independent they were not directly compared with each other (see is it valid below). In the lateral direction the position was determined from the anterior portal and the two lateral SSDs. The patient position was defined as the pelvic bony frame since the position of this would directly affect each of the measurement methods used. All measurements were normalised to the average measured with that method for that patient, so that only random variations were analysed. Clatterbridge Centre for Oncology John Sage, Tanja Wolff, Lynette Cassapi, Helen Mayles, Isabel Syndikus, Philip Mayles NHS Trust Separating patient daily set up error from measurement uncertainty in the analysis of portal images. Introduction : Current recommendations are that each centre should determine the local patient position variability using repeat portal image analysis for all clinical sites and set up techniques [1]. With such measurements it is important to be aware that the portal image analysis is not perfect and will itself have a random error, which will contribute to the variations determined. This work provides a method to measure the random variation in patient position in a way which is not affected by the inherent measurement uncertainty of the portal image analysis. The method also provides an estimate of the measurement uncertainty. Results : The precision of portal image verification varied from 1.3mm (1SD) for superior-inferior measurement from the anterior portal to 2.0mm (1SD) for anterior-posterior measurement from the lateral portal. Corrected patient position reproducibility was good in the superior-inferior and lateral axes with standard deviations of 1.4mm and 1.5mm respectively. The corrected patient position reproducibility in the anterior-posterior axis was 2.6mm (1SD) when set using the lateral tattoos. The precision of the digital couch reading for determining isocenter height was found to be 1.9mm (1SD). Below are shown example scatter plots of the agreement between different measurement techniques and the results of the full analysis. Analysis : Consider a measurement which is to be made on a patient with average position P and a variation around that position of  P. If the measurement has an associated error of  U then the result of the measurement (M) can be written as M =P+  P+  U For a series of measurements the standard deviation of M will be the quadrature sum of the positional variation and the measurement uncertainty.  M 2 =  U 2 +  P 2 Without further information it is impossible to know what proportion of  M is genuine patient position variation and what proportion is measurement uncertainty. In order to gain additional information is is possible to use an alternative, independent measurement of the patient position. With two measurement sources it is possible to examine the difference between the two measurements. The difference between the two measurements is purely a function of the two sources of measurement uncertainty. Measurement AM a = P +  P +  U a  M,a 2 =  U,a 2 +  P 2 Measurement BM b = P +  P +  U b  M,b 2 =  U,b 2 +  P 2 DifferenceM a - M b =  U a -  U b  (M,a-M,b) 2 =  U,a 2 +  U,b 2 There are now three equations and three unknowns, allowing the patient position variation to be determined independently to the measurement uncertainty. The same method can be used with more then two measurement methods using matrix algebra. Uncorrelated, suggesting small  P but larger  U Correlated, suggesting high  P but smaller  U Tight distribution, small  P and  U