The Radiation Oncology Feed For radiation oncologists · physicists · RTTs · dosimetrists
technical-relevance pancreatic cbct deep-learning

Deep learning reconstructed one-minute gated CBCT for pancreatic radiotherapy

Deep-learning reconstruction produced pancreatic gated cone-beam computed tomography comparable with conventional imaging while reducing estimated acquisition time and imaging dose.

RadOncToday · ·1 min read

KEY POINTS

  • The study included 156 clinical gated cone-beam computed tomography scans from 15 patients receiving respiratory-gated pancreatic radiotherapy.
  • Nonstop gated acquisitions were emulated by alternate-cycle downsampling of existing gated projection data and reconstructed using a dual-domain convolutional neural network.
  • The proposed technique would reduce half-fan acquisition to approximately one minute and full-fan acquisition to 33 seconds, compared with conventional scans lasting 2–8 minutes.
  • Estimated imaging dose was reduced by more than 40%.
  • Deep-learning reconstructions preserved soft-tissue contrast and anatomical detail, with the projection-domain network providing most of the improvement.

CLINICAL TAKEAWAY

Faster gated imaging could shorten pancreatic treatment sessions and reduce the opportunity for baseline drift or gastrointestinal anatomical change. Clinical readiness remains unproven because the nonstop scans were simulated from conventional acquisitions and were not used for prospective setup decisions.

SOURCE

Physics and Imaging in Radiation Oncology