1.  Introduction

The aim of this project is to provide a computer program to aid the routine comparison of portal and simulator radiographs for checking the accuracy of patient and beam setups in radiotherapy. Portal radiographs taken on a high energy treatment machine lack the detail of simulator films taken using much lower energy X-rays and require some image processing to enhance the visible features. In addition the two sets of films usually have different magnifications, rotations and translations and will require registering before a comparison can be made.

The goal of radiotherapy is to deliver a prescribed dose of radiation to a tumour as accurately as possible while minimizing the dose to normal surrounding tissues. Therefore, keeping the irradiated volume to a minimum is desirable. This is accomplished by keeping the treatment volume as close as possible to the tumour volume. However, this has to be balanced against the possibility of undertreating cancerous tissue due to positional uncertainties.

Studies of portal images have shown that variations in a patient setup greater than 1 cm are common (Byhardt, Cox ,Hornburgh & Liermann 1978, Rabionowitz, Broomberg, Goitein, McCathy & Leong 1985, Ding Shalev & Guchev 1993). Positional errors of this size mean that the full benefits from improvements in treatment planning by using computers for dose calculation and CT scans for tumour localization are not being realized. It is increasingly becoming the case that the positional accuracy to which the dose is delivered is the most significant and avoidable factor in treatment complications and failures (Williams & Thwaites 1993). It has been theoretically estimated that reductions in cure rates due to positional errors could be as high as 20% (Goitein 1975, McParland 1993).

Unfortunately the positional accuracy to which the prescribed dose is delivered is also the most difficult to control. This relates to the difficulties involved in setting up the patient, accurately and reproducibly from day to day. The problem of accurately setting up patients has been understood for some time (Marks, Hausm Sutton & Griem 1976), and it is now common practice to take portal films during a course of external beam radiotherapy as part of a complete quality assurance programme. The portal films can then be compared with the simulator film to ensure that patient alignment matches that at the simulation stage, and remains consistent throughout the treatment.

The use of portal images also provides a valuable tool for checking the placement of shielding blocks.

A significant difficulty encountered in the use of portal images for treatment verification is the low contrast that they exhibit. Bone, producing contrast of 18% at photon energies of 50keV, will only show a 2% contrast when imaged at an energy of 6MeV (Williams & Thwaites 1993). In addition at this higher treatment energy, there are large amounts of scatter further reducing contrast in the image. Unfortunately suitable grids are not available to overcome this. The resultant poor quality of portal images makes their comparison with the prescription (simulator) images a difficult and time-consuming process. A further complication is that the two images are usually at different magnifications, and commonly have rotations and translations with respect to each other, often as a result of film placement as much as than patient setup. These factors combine generally to limit the use of portal images in the modern hospital environment to one check film on the first fraction, which is far from the ideal of verification at every treatment.

Advances in megavoltage imaging systems mean that treatment machines can now be fitted with on-line portal imaging, allowing the acquisition of digital portal images in the first few seconds of treatment. If techniques for the rapid comparison of these images to those acquired at the simulation stage can be developed, a potential for the pro-active detection and correcting of poor patient alignment exists. In this scenario, at the beginning of each treatment fraction, a portal image will be taken and compared with the simulator image, and any problems in the patient setup can then be resolved before the radiation dose is delivered. This is favorable to the current practice of retrospectively correcting the patient setup based on a portal image taken at a previous treatment fraction, with the possibility of re-simulation in more difficult cases. The improvements in patient setups this would bring may well lead to a reduction in treatment complications and failures and clearly if this were the case would represent a major step forward in accurate patient setups.

For the time being, the requirements of the program are to display portal and simulator images digitized on a scanner. After image processing to enhance the contrast, registration will be carried out. This will correct differences in magnification and bring the two images into alignment, thus making comparisons between the images much simpler. Finally a facility to compare the two images either qualitatively or quantitatively will be required.