Radiation therapy dose prediction for left-sided breast cancers using two-dimensional and three-dimensional deep...

Area of Science:

• Medical Physics
• Radiotherapy
• Artificial Intelligence

Background:

• Accurate dose distribution is crucial for effective radiotherapy.
• Deep learning offers potential for automating and improving treatment planning.

Purpose of the Study:

• Develop a deep learning model for generating clinically acceptable dose distributions for 3D-CRT in left-sided breast cancers.
• Compare the performance of 2D versus 3D anatomical data inputs for dose prediction.

Main Methods:

• Trained two U-net based deep learning models (2D and 3D) to predict dose distribution.
• Utilized patient CT scans, organ-at-risk (OAR) masks, and dose prescription as input.
• Validated models using 5-fold cross-validation on 120 patients and tested on 25 patients.

Main Results:

• Both 2D and 3D models produced clinically acceptable dose distributions.
• The 3D model demonstrated superior performance compared to the 2D model.
• Average dose differences for mean dose were within 0.02% of prescription; V20 values were comparable to clinical plans.

Conclusions:

• Deep learning models can be clinically implemented to generate patient-specific dose distributions.
• These models can serve as a reference for optimal treatment planning.
• The approach enhances plan quality efficiently without hindering the planning process.