Black and white in accuracy assessment of megavoltage images: the medical decision is often grey

Abstract

Using different criteria for acceptance of the portal film taken at the first treatment session, a comparison was made of the relevance of the information obtained from such a single check for the subsequent irradiations. A total number of 234 verification films have been taken on 29 fields for 27 head and neck patients. Patients were immobilised with individual plastic masks fixed to the couch and treated on a 6-MV linac fitted with an automatic verification system. Field alignment was checked with a measurement in the anteroposterior (AP) and craniocaudal (CC) direction on each film. Referring to the simulated field, this group of patients was treated with excellent average precision (mean, −0.7 mm) and reasonable spread (s = 5 mm). The percentage of ‘large’ deviations (≥6 mm) occurring during the whole treatment course is proportional to the upper limit of deviation accepted in the first assessed field (for an upper limit <6 mm): it goes progressively up from 5% (AP-CC direction) to 17% (AP) and 13% (CC) for accepted magnitudes of deviation going from 2 mm to 6 mm in the first film. As the reproducibility of the different treatment series (s = 2 mm) is independent of the upper level of error accepted on the first film, this means that errors are mainly systematic errors coming from the transfer of the simulation unit to the treatment unit. Precision in a series of set-ups is always expressed by a Gaussian curve. A control port film at the first session is most frequently representative for the median value of the whole series and subsequent set-ups always show a spread of values. It is therefore important to choose an upper limit of the acceptable deviation on the first measurement which is small enough, as compared with the ‘unacceptable deviation’, to be sure that most of the subsequent set-ups can also be considered as correct. The acceptance limit of a single first image should therefore be directly linked to the known precision of a technique (determined by the standard deviation of the reproducibility distribution) and to the clinical situation involved. It should not be limited to a ‘black’ or ‘white’ decision with a single numerical value for different treatment modalities.