Deep learning-based dose prediction to enhance planning efficiency in cervical brachytherapy with hybrid applicators

Area of Science:

• Medical Physics
• Oncology
• Radiotherapy Technology

Background:

• Cervical cancer brachytherapy often utilizes hybrid applicators.
• Accurate dose prediction is crucial for effective treatment and minimizing side effects.

Purpose of the Study:

• To develop and validate a dose prediction model for hybrid applicators in cervical cancer brachytherapy.
• To assess the model's accuracy and clinical applicability.

Main Methods:

• Utilized 216 treatment plans across three hybrid applicator types: tandem-ovoids with needles, tandem-ring with needles, and tandem-cylinder with needles.
• Trained and tested a predictive model using distinct datasets.
• Validated the model using Mean Absolute Error (MAE) and Dice Similarity Coefficient (DSC) on test and clinical evaluation datasets.

Main Results:

• The model achieved a voxel-wise MAE of 0.45 ± 0.27 Gy.
• No statistically significant differences (p > 0.05) were found in key dose-volume histogram parameters compared to clinical plans.
• DSC values ranged from 0.74 to 0.77, with minor dosimetric differences observed for HR-CTV and bladder.

Conclusions:

• The developed dose prediction model shows significant potential for hybrid applicator brachytherapy.
• The model can guide dose delivery optimization, streamline planning, and enhance consistency in adaptive workflows.