Fluence Map Prediction Using Deep Learning Models - Direct Plan Generation for Pancreas Stereotactic Body Radiation...

Area of Science:

• Medical Physics
• Radiotherapy
• Artificial Intelligence

Background:

• Pancreas stereotactic body radiation therapy (SBRT) planning is complex and time-consuming.
• Current methods require extensive manual input and iterative optimization.

Purpose of the Study:

• To develop a novel deep learning framework for automated pancreas SBRT treatment planning.
• To generate clinical-quality plans by directly predicting fluence maps using convolutional neural networks (CNNs).

Main Methods:

• A two-CNN framework was developed: one for field-dose prediction and another for fluence map prediction.
• The framework utilizes patient anatomy and planning images to predict intensity-modulated radiation therapy (IMRT) fluence maps.
• Model-predicted plans were compared against benchmark plans from 100 retrospective pancreas SBRT cases.

Main Results:

• The deep learning framework achieved an average fluence map prediction time of 7.1 seconds per patient.
• Model-predicted plans demonstrated comparable dosimetric endpoints and deliverability to benchmark plans.
• Absolute dose differences for target and organs-at-risk were minimal, with high gamma indices.

Conclusions:

• A novel deep learning framework for pancreas SBRT planning was successfully developed.
• The framework bypasses the inverse optimization process, generating deliverable plans rapidly.
• This approach has the potential to revolutionize treatment planning efficiency and accessibility.