Automatic lung dose painting for functional lung avoidance radiotherapy through multi-modality-guided dose...

Area of Science:

• Medical Physics
• Radiotherapy
• Medical Imaging

Background:

• Functional Lung Avoidance Radiotherapy (FLART) aims to minimize radiation dose to healthy lung tissue.
• Accurate prediction of dose distribution is crucial for effective FLART planning.
• Current auto-planning methods may not fully utilize voxel-wise lung function information.

Purpose of the Study:

• To develop a multi-modality-guided dose prediction (MMDP)-based auto-planning algorithm for FLART.
• To leverage voxel-wise lung function images for enhanced dose prediction and plan generation.
• To improve the efficiency, consistency, and quality of FLART planning.

Main Methods:

• Developed a novel MMDP model extracting complementary features from multi-modality images.
• Implemented an instance-weighting anatomy-to-function training strategy to improve prediction accuracy.
• Utilized a function-guided voxelwise dose mimicking algorithm to create MMDP-FLART plans.
• Validated the algorithm on retrospective and prospective patient data with SPECT ventilation (V) and perfusion (Q) images.

Main Results:

• MMDP achieved accurate dose predictions with median errors within ±1Gy/±1% for DVH metrics.
• The MMDP model and training strategy significantly reduced prediction errors for functionally weighted mean lung dose (fMLD).
• MMDP-FLART plans demonstrated significant reductions in fMLD compared to conventional radiotherapy (ConvRT) plans.
• MMDP-FLART plans showed comparable or lower fMLD than manual FLART plans, with reduced dose to organs at risk.

Conclusions:

• The MMDP model with instance-weighting anatomy-to-function training enables accurate dose prediction for FLART.
• The MMDP-based auto-planning algorithm effectively generates FLART plans using voxel-wise lung function data.
• This approach shows potential to enhance FLART planning efficiency, consistency, and overall quality.