Predicting the 3-Dimensional Dose Distribution of Multilesion Lung Stereotactic Ablative Radiation Therapy With...

Area of Science:

• Medical Physics
• Radiation Oncology
• Artificial Intelligence in Medicine

Background:

• Stereotactic ablative radiotherapy (SABR) is increasingly used for lung metastases.
• Treating multiple lung lesions presents challenges in optimizing dose and fractionation.
• Complex plans involve diverse lesions, heterogeneous prescriptions, and interacting dose distributions.

Purpose of the Study:

• To develop and evaluate a generative adversarial network (GAN) for real-time dosimetry prediction.
• To address the complexities of multilesion lung SABR therapy.
• To balance local failure and treatment toxicity in complex cases.

Main Methods:

• A GAN was trained using patient data from a tertiary center.
• Input included CT scans, organs at risk, target structures, and dose fall-off estimates.
• Model performance was evaluated using voxel-voxel, DVH, and conformity metrics, with out-of-sample validation.

Main Results:

• The GAN achieved mean absolute dosimetry differences <3 Gy and gamma pass fractions >90% for OARs.
• Lung V20 and CV14 differences were 1.40% ± 0.99% and 75.8 ± 42.0 cc.
• Out-of-sample validation confirmed model accuracy and low variance in predictions.

Conclusions:

• A GAN effectively predicts 3-D dosimetry for complex multilesion lung SABR.
• Rapid dosimetry prediction aids in assessing treatment feasibility.
• This tool can explore dosimetric variations from different prescriptions.