ESTRO 2024 - Abstract Book

S3933

Physics - Image acquisition and processing

ESTRO 2024

Conclusion:

This study investigated the use of deep learning super-resolution to reduce CBCT imaging dose. The results demonstrated that ESRGAN is capable of generating images with a similar resolution and noise characteristic as currently used CBCTs, but from low-resolution projections acquired with 25% of the original imaging dose. The deep learning super-resolution outperformed conventional bilinear up-sampling in image similarity, noise and resolution metrics. Currently low-resolution projections were simulated, but the above outcomes motivate experimental studies with real low-dose acquisition of phantoms and eventually patients.