Magnetic resonance imaging-based proton dose calculation for pelvic tumors using deep learning

Comprehensive Summary

In this study, two deep learning models were compared in their ability to perform MRI-only proton therapy dose calculations in pelvic cancer patients. The first model was a two-step pipeline that converts the MRI images to a synthetic CT, and then makes predictions on proton dosage. The second model was a direct pipeline, that converted MRI image to proton dose prediction directly. The study aimed to evaluate how MRI intensity distortions can affect the accuracy of dose prediction. These models were evaluated using average relative error (ARE) and gamma pass rate (3mm/3%), comparing the performance of the pipelines to Monte Carlo-simulated doses. The results show that while the two-step pipeline had lower dose errors, both pipelines displayed high accuracy, with gamma pass rates exceeding 99%. Therefore MRI only proton dose caculation is feasible for pelvic cancer patients.

Outcomes and Implications

This study demonstrates that the integration of deep learning models into MRI-only proton dose calculation is feasible for pelvic cancer patients. This could reduce reliance on CT, therefore lowering radiation exposure.