Fast Deformable Image Registration for Real-Time Target Tracking During Radiation Therapy Using Cine MRI and Deep...

Area of Science:

• Medical Imaging
• Radiotherapy
• Machine Learning

Background:

• Accurate target tracking is crucial in radiation therapy to ensure precise dose delivery.
• Conventional image registration methods can be computationally intensive and may lack the speed required for real-time applications.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) model for fast deformable image registration using cine MRI during radiation therapy.
• To compare the DL model's performance against conventional methods for real-time target tracking.

Main Methods:

• A DL model was developed to process pairs of 2D sagittal cine MRI images, outputting a motion vector field (MVF) for image alignment.
• The model was trained and evaluated on cine MRI data from patients undergoing treatment for abdominal and thoracic tumors.
• Performance was assessed using registration error, MVF stability, and computation time, comparing DL against affine, b-spline, and demons registration.

Main Results:

• The DL model demonstrated superior performance in reducing registration error compared to conventional methods (RMSE: 0.032 for DL vs. 0.036-0.067 for others).
• DL registration achieved significantly faster computation times per frame (8 ms) than affine, b-spline, and demons methods.
• The DL model exhibited comparable spatial MVF stability and improved temporal consistency over deformable methods.

Conclusions:

• Deep learning-based image registration effectively utilizes large-scale MR cine data.
• The DL approach offers a promising solution for real-time deformable motion estimation in radiation therapy, outperforming traditional techniques.