An energy fluence-convolution model for amorphous silicon EPID dose prediction.

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Image Processing

Background:

• Amorphous silicon electronic portal imaging devices (a-Si EPIDs) are crucial for verifying radiation delivery in radiotherapy.
• Accurate dose prediction in EPIDs is essential for quality assurance, especially for complex treatments like intensity modulated radiotherapy (IMRT).
• Existing models may not fully account for multileaf collimator (MLC) effects and backscatter, impacting prediction accuracy.

Purpose of the Study:

• To develop and validate an a-Si EPID dose prediction model integrated with the Pinnacle treatment planning system.
• To incorporate high-resolution energy fluence and MLC leaf effects into the EPID dose prediction.
• To improve the accuracy of EPID image prediction for IMRT fields by addressing backscatter and MLC transmission.

Main Methods:

• Developed a high-resolution (< 1 mm) energy fluence matrix model to simulate MLC leaf effects.
• Calculated primary dose using experimentally derived, radially dependent EPID interaction coefficients for open and MLC-transmitted components.
• Convolved primary dose with a spatially invariant kernel accounting for radiative dose, backscatter, and optical glare, optimized for penumbra and scatter factor prediction.
• Implemented an EPID calibration method to mitigate nonuniform E-arm backscatter effects.

Main Results:

• The model accurately predicted EPID images for jaw and MLC-defined open fields and head and neck IMRT fields.
• For head and neck IMRT fields, 97.6% of points passed with a 2%/2 mm gamma index, and 99.4% passed with 3%/3 mm.
• Removing E-arm backscatter from the pixel sensitivity correction matrix improved the 2%/2 mm pass score from 96.0% to 97.6%.

Conclusions:

• The developed a-Si EPID dose prediction model accurately incorporates MLC leaf transmission and EPID response to dose components.
• The model effectively predicts EPID images for IMRT fields, demonstrating high agreement with measured data.
• The proposed calibration method significantly reduces the impact of nonuniform E-arm backscatter, enhancing overall prediction accuracy.