Monte Carlo Dose Calculation Using MRI Based Synthetic CT Generated by Fully Convolutional Neural Network for Gamma...

Area of Science:

• Medical Physics
• Radiosurgery Technology
• Artificial Intelligence in Medicine

Background:

• Stereotactic radiosurgery (SRS) typically relies on CT scans for accurate radiation dose calculation.
• Integrating MRI data with deep learning offers a potential alternative for generating synthetic CT (sCT) images.
• This approach could enhance treatment planning for Gamma Knife (GK) SRS by leveraging the superior soft-tissue contrast of MRI.

Purpose of the Study:

• To evaluate the dosimetric accuracy of synthetic CT (sCT) images generated using a deep learning algorithm from MRI data for Gamma Knife (GK) stereotactic radiosurgery (SRS).
• To assess the performance of a deep convolutional neural network in translating MR images to sCT.
• To quantify the impact of sCT on radiation dose calculations compared to conventional CT scans.

Main Methods:

• A deep convolutional neural network (encoder-decoder architecture) was trained to generate sCT from T1-weighted MR images.
• Thirty patients undergoing GK SRS were retrospectively analyzed, with MR and CT datasets preprocessed and normalized.
• Monte Carlo (MC) simulations were performed for dose calculations using both true CT and generated sCT images to assess dosimetric accuracy.

Main Results:

• The deep learning model achieved a Mean Absolute Error (MAE) of 86.6 ± 34.1 HU and a Mean Squared Error (MSE) of 160.9 ± 32.8.
• A mean Dice similarity coefficient of 0.82 ± 0.05 was obtained for regions with HU > 200.
• Dosimetric analysis showed a minimal difference of 1.1% in the D95 dose-volume parameter when using sCT with a CT-to-density table, compared to 4.9% without it.

Conclusions:

• Deep learning-based sCT generation from MR images is a viable method for GK SRS.
• The generated sCT images demonstrate high fidelity and accuracy, suitable for treatment planning.
• Utilizing sCT with an appropriate CT-to-density conversion significantly preserves dosimetric accuracy in SRS, potentially reducing the need for CT scans.