Multislice motion modeling for MRI-guided radiotherapy gating.

Area of Science:

• Medical Imaging
• Radiotherapy
• Computational Anatomy

Background:

• On-board MRI enhances real-time target tracking in radiotherapy.
• Current multislice and volumetric MRI techniques have frame rate limitations causing latency.
• This latency can lead to inaccuracies in turning off the radiation beam when the target moves.

Purpose of the Study:

• To develop a technique for continuous volumetric tissue motion estimation using motion models.
• To apply these models for multislice target visualization during radiotherapy.
• To enable out-of-slice motion estimation for improved MRI-guided radiotherapy.

Main Methods:

• Acquired images at 3 frames/sec across ten sagittal slices using a 0.35 T MRI-guided radiotherapy system.
• Extended a 5D linear motion model to include multiple slices, using respiratory bellows signals for simultaneous motion estimation.
• Utilized manifold learning (Local Linear Embedding) for motion modeling and developed the multi-slice Groupwise Alignment (mSGA) technique for out-of-slice motion estimation.

Main Results:

• The mSGA model achieved higher gating decision accuracy (96.8%) and positive predictive value (96.1%) compared to the 5D model (93.7% and 92.8%).
• mSGA reduced the median and 95th percentile distance between model and ground truth target centroids to 0.58 mm and 1.49 mm, respectively.
• These improvements were statistically significant across all evaluated metrics.

Conclusions:

• The proposed out-of-slice target motion estimation techniques show accuracy suitable for clinical application.
• The mSGA model demonstrates potential for higher accuracy in motion estimation.
• The external surrogate-based model facilitates unbiased in vivo accuracy evaluation.