MVCT Reconstruction from Irregular EPID Projections Using Deconvolution Method

Accurate patient positioning is critical in radiotherapy, but conventional CBCT-based verification faces challenges such as mechanical isocenter inconsistencies, increased radiation dose, and prolonged treatment time. This study utilizes the backprojection based deconvolution algorithm to reconstruct megavolt CT (MVCT) images from irregular EPID treatment beam projections without the need for additional radiation. This method uses convolutional kernels to iteratively remove artifacts introduced by direct backprojection. To compensate for incomplete data from irregular fields, TV regularization, L0 regularization, and guided image filtering (GIF) are incorporated. Experimental results on a chest phantom show that deconvolution alone provides limited quality, TV regularization enhances structural fidelity, and the combination of L0 regularization and GIF achieves the best performance (PSNR = 44.96 dB, SSIM = 0.924). These results demonstrate that our framework enables high-quality MVCT reconstruction from non-ideal EPID data, offering a promising strategy for dose-free positioning verification in radiotherapy.