Purpose: We evaluated the performance of a new dosimetry module in the LC250 scanning liquid-filled ionization chamber (SLIC) electronic portal imaging device (EPID) for intensity-modulated radiotherapy (IMRT) verification. This module permits one to convert EPID readings to two-dimensional (2D) maps of IMRT dose rate in real time, and to integrate them over time to produce a profile of accumulated dose for treatment verification. Methods and Materials: The EPID was calibrated using an iterative procedure, from which a lookup table for dose integration was generated and transferred to the image-acquisition hardware. To evaluate the EPID's integration capability, we investigated the linearity of imaging time (vs. monitor unit [MU]) and integrated dose (vs. planned dose) for static and IMRT fields, in both standard (similar to2.7 s/image) and fast (similar to1 s/image) synchronous acquisition modes (S- and F-modes). We also compared the EPID-measured profiles with that measured using film and ionization chamber, or calculated from the treatment planning system. For the EPID's patient dose verification capability, we compared the integrated central-axis (CAX) dose with the planned dose for 25 prostate IMRT fields. We also compared the measured relative profiles with the planned ones using a linear regression model, which returns an index or (root mean squared error) for the goodness of fit. We identified errors that are either associated with the timing of the EPID-start delay and end truncation, or with the integration process-detector memory effects (decrease in detector's sensitivity with time during the fast continuous acquisition) and beam hold-off effects (the withholding of linac beam pulses when multileaf collimator leaves are not in the correct positions). The CAX doses of static fields were corrected using the ratio of the irradiation time to the imaging time. A linear decay model was proposed to correct the detector memory effect. To investigate the beam hold-off effect, we verified the relative profiles of a five-field prostate 1MRT plan for five different MU settings, and correlated the goodness of fit with the percent of beam hold-off. Results: The imaging time is linearly proportional to the given NW with a slope of 0.250 MU/s (ideal slope is 0.250 MU/s) and a R-2 = 1.0. Although the R-2 of the linearity for the measured vs. planned dose is 1.0 for both modes, only the slopes for the S-mode are within 3% of unity. The slopes for the F-mode deviate from unity due to detector memory effects, and are accurately corrected using the linear decay model. The EPED measured profiles agree well (within 2.0%) with the planned dose and profiles for both modes. For the CAX dose of the 25 IMRT fields, the S-mode is within 2% of the planned dose, whereas the F-mode is off significantly (>3%) if not corrected for detector memory effects. For the relative profile verification, lower MU always produces higher sigma, for the same mode. The F-mode is more accurate than the S-mode for the same MU; however, the improvement is not proportional to the difference in imaging speed. Analysis of the correlation of the goodness of fit with the percent of beam hold-off indicates that the accuracy of profile verification for the F-mode is predominantly determined by the beam hold-off effect for lower MU. Conclusion: The S-mode of LC250 combined with a large MU can be used for the pretreatment verification of IMRT beam delivery with a significant reduction of processing time and computer resources in comparison to off-line processing. Real-time verification during treatment requires the F-mode. Although the detector memory effects encountered in the F-mode can be compensated using the proposed linear decay model, sufficient accuracy for real-time verification requires a resolution of the beam hold-off problem. (C) 2003 Elsevier Science Inc.
