Simulation-free workflow for lattice radiation therapy using deep learning predicted synthetic computed tomography: A...

Area of Science:

• Medical Physics
• Radiotherapy
• Artificial Intelligence in Medicine

Background:

• Lattice radiation therapy (LRT) is a palliative treatment for bulky tumors, often requiring complex planning.
• Expediting treatment initiation is crucial for eligible patients undergoing LRT.

Purpose of the Study:

• To develop a simulation-free workflow for volumetric modulated arc therapy (VMAT)-based LRT planning.
• To utilize deep learning-predicted synthetic CT (sCT) to expedite treatment initiation.

Main Methods:

• Trained two 3D U-Net deep learning models to generate sCT from diagnostic CTs (dCT) for thoracic and abdominal regions.
• Validated models on an independent dataset using image similarity metrics (MAE, SSIM).
• Generated VMAT-LRT plans on sCT and planning CT (pCT) for dosimetric comparison using DVH metrics.

Main Results:

• The sCT prediction model achieved high image similarity to pCT (MAE: 38.93±14.79 HU, SSIM: 0.92±0.05 for thoracic; MAE: 73.60±22.90 HU, SSIM: 0.90±0.03 for abdominal).
• No statistically significant differences in DVH parameters for organs-at-risk and target volumes between sCT and pCT plans, except for Dmin and D10%.

Conclusions:

• Deep learning-predicted sCT demonstrates high image similarity and adequate dose agreement with pCT.
• This study serves as a proof-of-concept for a simulation-free VMAT-LRT planning workflow using DL-predicted sCT.