A regression-based approach to compute the pixels sensitivity map of linear accelerator portal imaging devices.

PURPOSE

To determine the pixel sensitivity map (PSM) for amorphous silicon electronic portal imaging devices (EPIDs) using a single flood field signal.

METHOD AND MATERIALS

A raw EPID signal results from the incident particle energy fluence, the inherent pixels response, and the background signal. In large open fields, particle energy fluence is a slow-varying signal that is locally considered spatially constant. Pixels response is a fast and abrupt varying behavior. The background signal is due to the EPID panel electronics, which is determined during radiation absence. To determine the PSM, after correcting for the background signal, we apply a model that captures the underlying smooth particle energy fluence-induced signal. This fluence signal-fitted model is then used to determine the PSM. Here, we use a polynomial-based regression surface model in both x and y dimensions. To validate the generated PSM, we measure beams and compute PSMs for multiple beam energies with and without flattening filters and for multiple source-to-imager distances. Since the PSM is a detector characteristic, it should be independent of those variables. We also intercompare measurements of fixed slit fields with the EPID being shifted between measurements.

RESULTS

The fluence signal of the flattening filter-free (FFF) beams was optimally modeled as a 12th degree polynomial surfaces, which had 0.1% residuals near the central axis. The 6 and 10 MV FFF PSMs were within ˜0.1%, and independent of the EPID SID, suggesting that the PSM is energy independent. The 6, 10, and 15 MV flattened-beam PSMs were well modeled as 12th degree polynomial surfaces, which were equivalent within ˜0.24% but differed from the FFF PSM by up to 0.5% near the beam central axis. Applying the FFF PSMs to the flattened-beam measurements reduced the central-axis deviation between the raw and corrected signal to 0.1%, confirming the PSM energy independence hypothesis. When the FFF PSM is utilized, output verification with shifted slit deliveries agreed within ˜0.5% for all beam energies, which is within the radiation delivery uncertainty of ˜0.57%.

CONCLUSION

PSM for MV EPIDs can be determined by separating out the slowly varying, well-behaved fluence signal from the pixel-to-pixel sensitivity variations. The quality of the PSM is found to be dependent on the quality of the surface fit, which is best for the 6 MV FFF beam measured at SID equal to 180 cm. Within fitting errors, the PSM is independent of beam energy for 6, 10, and 15 MV beams with and without flattening filters. The PSM generation does not require shifting the EPID panel nor multiple EPID panel irradiations and should be usable for linacs with fixed geometry EPIDs.