Megavoltage CT enhancement for cervical cancer tomotherapy using a generative adversarial network with deformable convolution and self‐attention

Key Points

• The aim is to enhance image quality and Hounsfield unit (HU) accuracy in cervical cancer imaging using a novel framework.
• Developed the SADC-EGAN framework incorporating deformable convolutions and self-attention.
• Generated kVCT-like images from MVCT scans.
• Evaluated improvements in image quality and HU accuracy.
• Significant enhancement in image quality was observed.
• HU accuracy improved, facilitating better clinical assessments.