Putting the beam profile back into flood-field corrected EPID images
Motivation
Clinical EPID images are usually automatically flood-field corrected: the signal in each pixel of each image is divided by the signal in the corresponding pixel in an image of a large, open field. This correction removes the effects of variations in both pixel sensitivity and beam intensity from the resulting images. Removal of the lateral variation of the beam intensity causes difficulties when the EPID images are processed for dosimetry because it results in the removal of the record of a real and possibly strong variation in delivered dose. Not all clinics and research centres have access to the resources (software, hardware) that are needed to avoid or automatically negate the flood-field correction. And sometimes total avoidance of the flood-field correction is undesirable, due to radical variations in EPID pixel sensitivity. It would therefore be useful to have a straightforward means to obtain EPID images that incorporate the varying beam profile, even when those images have been automatically flood-field corrected. Here is a suggested (simple but simplistic) method…
Method
(Note that, in the results shown here, open data points represent results from Monte Carlo simulations and filled data points represent results from experimental EPID images.)
1. Run a series of Monte Carlo simulations of your linear accelerator and EPID, with the beam attenuated (and hardened and scattered) by a series of isocentrically-positioned rectilinear phantoms, of various thicknesses (preferably matching the dimensions of phantoms to which you have access).
2. Identify a phantom thickness which results in a flat field being measured by the simulated EPID.
3. Experimentally, obtain an EPID image (which has been automatically flood-field corrected) of the phantom identified in step 2. Call this image Iflat. (Notice that it isn’t flat.)
4. Now forget about your Monte Carlo results.
5. For each experimental image (n) that you wish to correct, do the following…
(a) Normalise to I0:
(b) Calculate difference between I0 and Iflat, as a ratio:
(c) Flood-field un-correct your experimental images, using the ratio:
6. If you want to, you can confirm that your experimental images are now “un-corrected” (ie. that they include the beam profile) by comparing them with never-corrected Monte Carlo simulation results. Example:
These examples indicate that the flood field un-correction is working and is necessary, for large fields (in the left hand example, the field size is 25cm x 25cm at the isocentre) and that the flood-field un-correction is not necessary for small central-axis fields (in the right hand example, the field size is 5cm x 5cm at the isocentre). It would be much more interesting to test this procedure using images (and CT data) from a clinical treatment, or at least from treatment-like fields delivered to a patient-like phantom.