ESTRO 2024 - Abstract Book

S4162

Physics - Intra-fraction motion management and real-time adaptive radiotherapy

ESTRO 2024

We optimized an image sequence to adhere to the following requirements: 1) < 2 seconds per frame to resolve respiratory and peristaltic motion, 2) 3D volume 300x75x300 mm 3 to image the full bowel, 3) visually good contrast of the bowel, 4) good response in image registration. Given the last requirement, we optimized the sequence synergistically with the outcomes of the image registration optimization (see below).

Image registration

For motion tracking we investigated several registration methods, that previously showed to be promising for radiotherapy purposes [2].

The image frames were registered in a consecutive fashion, in which each registered frame was used as reference for the next step. Every 5th acquired frame was used for registration, resulting in 100 steps for 15 min. imaging.

Validation

An MRI scan with optimized setting was acquired during 15 minutes on a male volunteer on a 1.5T MRL (Unity, Elekta AB), without bowel preparation or instruction regarding respiration.

The quality of the estimated deformations was evaluated using two criteria.

1. The total temporal variation of image intensity (TIV), evaluating the alignment between frames in a time series. 2. The inverse consistency (IC) of the deformation vector fields, which provides information concerning the precision of the estimated deformations.

Both were evaluated on the complete bowel bag mask.

Results:

The following sequence gave the best results in the subsequent image registration: a coronal balanced Fast Field Echo (FFE) sequence with TR = 4.6ms, voxel size = 0.98 x 2.25 x 0.98 mm 3 and dynamic scan time 1.8s.

For image registration the Cornelius-Kanade [3] algorithm was the most optimal based on quality metrics and visual inspection. This algorithm is a derivation from the Horn-Schunck algorithm, with increased robustness to image intensity variations not attributed to motion (e.g. artifacts and transient image features), which are common in bowel imaging.