Deep learning dose prediction for hybrid applicators improves cervical brachytherapy planning consistency

A deep learning dose prediction model for cervical brachytherapy hybrid applicators achieved voxel-wise MAE of 0.45±0.27 Gy, with no significant differences in key DVH parameters (p>0.05) versus manual plans, in a retrospective study of 216 plans.

• Retrospective study of 216 treatment plans (161 training, 40 testing, 15 clinical evaluation) using three hybrid applicator types: tandem/ovoids, tandem/ring, and tandem/cylinder with interstitial needles.
• Small dosimetric differences were observed for HR-CTV D98%, HR-CTV D90%, and bladder D2cc compared to manually generated plans.
• Limitation: single-center, retrospective design with a small clinical evaluation set; no external validation reported.

RadPigeon summaries are original and for information only. They are not clinical advice.