Improving Radiotherapy Plan Quality for Nasopharyngeal Carcinoma With Enhanced UNet Dose Prediction

Area of Science:

• Medical Physics
• Radiation Oncology
• Artificial Intelligence in Medicine

Background:

• Individualized dose prediction is crucial for optimizing radiation therapy.
• The DESIRE model, an enhanced UNet-based system, is developed for nasopharyngeal carcinoma (NPC) patients undergoing volumetric modulated arc therapy.
• This study evaluates DESIRE's impact on improving radiation treatment plan quality.

Purpose of the Study:

• To assess the effectiveness of the DESIRE model in predicting radiation dose distributions for NPC patients.
• To evaluate the impact of integrating DESIRE into the treatment planning process.
• To determine if DESIRE improves plan quality compared to standard methods.

Main Methods:

• A retrospective study of 131 NPC patients was conducted.
• The DESIRE model predicted dose distributions, with discrepancies from ground truth (GT) quantified using dosimetric metrics and gamma pass rates.
• Junior physicians utilized DESIRE predictions for treatment planning, and their plans were compared to GT.

Main Results:

• DESIRE's predicted dose metrics closely aligned with GT, with mean differences < 1 Gy for organs at risk (OARs).
• While PTV metrics showed differences, mean differences for key dosimetric parameters did not exceed 1 Gy.
• Physician plans assisted by DESIRE were comparable to GT, with improved conformity and homogeneity indices for PTV70.

Conclusions:

• The DESIRE model demonstrates potential for accurate patient-specific dose prediction in radiation oncology.
• Integration of DESIRE enhances junior physicians' treatment planning capabilities.
• DESIRE contributes to improved radiation therapy plan quality for NPC patients.