An MR-only acquisition and artificial intelligence based image-processing protocol for photon and proton therapy...

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Image Processing

Background:

• MR-guided radiation therapy is increasingly feasible, but open, low-field MRI scanners have limited fields of view, truncating anatomy and impacting MR to synthetically generated CT (sCT) conversion.
• This limitation hinders the clinical application of MR-only workflows, particularly for proton therapy.

Purpose of the Study:

• To develop and validate an acquisition protocol and MR image stitching technique to overcome field-of-view restrictions in open MR scanners.
• To enable accurate sCT generation for MR-only photon and proton therapy.

Main Methods:

• An enhanced imaging protocol with repeated scans and table movement was used on 12 prostate cancer patients with an open 0.35T scanner.
• Stitched MR images were used for delineation, converted to sCT via a neural network, and used for inverse planning of VMAT and SFUD photon/proton plans.
• Plans were recalculated on CT images, and comparisons were made using dose metrics (D2%, D50%, V70Gy, D98%) and Dice coefficients for structure delineation accuracy.

Main Results:

• Maximum surface distance analysis showed average lateral uncertainties of 1-3mm between stitched MRI and CT.
• The sCT conversion demonstrated high accuracy, with Dice coefficients of 92-100% for femoral bones and the whole body.
• Dose comparison revealed uncertainties below 1% (deformed CT) and 2% (rigidly registered CT) in the CTV for both photon and proton plans.

Conclusions:

• The developed acquisition protocol and sCT generation method are well-accepted for MR-guided photon and proton therapy.
• This approach effectively addresses the limited field-of-view issue of open MR scanners, expanding capabilities for MR-guided proton therapy.