Improving Image Quality for Radiotherapy Treatment Planning: A Study of Image Acquisition, Processing, and Calibration Using CycleGAN Model

This article explores image acquisition and processing techniques used for radiotherapy treatment planning in NPC patients. The study is approved by the ethics committee of Fujian Cancer Hospital and all patients have provided written informed consent. CT and CBCT imaging techniques are used to obtain high-quality images that are registered and resampled using open-source software 3D-Slicer. Binary masks are created to avoid non-anatomical structures during training, and a CycleGAN model is used for image generation. The generator contains an encoder, conversion layer, and decoder, while the discriminator is a binary network trained with Adam optimizer. The study evaluates the improvement of image quality using Mean Absolute Error (MAE), gamma pass rates, dosimetric parameters, and side-by-side comparisons. The findings are statistically significant and offer insights for radiotherapy treatment planning in NPC patients.