Session Item

Monday
November 30
08:00 - 08:40
Physics Stream 1
Validation and commissioning of AI contouring tools
3020
Teaching Lecture
Physics
11:18 - 11:26
Quantifying intra-fractional gastric wall motion for MR-guided radiotherapy
PH-0127

Abstract

Quantifying intra-fractional gastric wall motion for MR-guided radiotherapy
Authors: Driever|, Theo(1)*[t.driever@nki.nl];van der Horst|, Astrid(2);Teuwen|, Jonas(1);Fast|, Martin(3);Sonke|, Jan-Jakob(1);
(1)Netherlands Cancer Institute, Radiation Oncology, Amsterdam, The Netherlands;(2)Academic Medical Center, Radiotherapy, Amsterdam, The Netherlands;(3)University Medical Center, Radiotherapy, Utrecht, The Netherlands;
Show Affiliations
Purpose or Objective

Pre-operative radiotherapy is a novel approach in gastric cancer management. The stomach, however, is an organ that is susceptible to various sources of geometric uncertainty such as respiration and peristalsis. Consequently, the stomach is not only subject to displacement but also deformation. The purpose of this study was to analyse the relative contribution of rigid displacement and non-rigid gastric wall deformation in coronal MR images after correcting for respiratory motion.

Material and Methods

Four healthy volunteers underwent an MRI examination on the Unity MR-linac (Elekta AB, Stockholm, Sweden), including a T2-weighted retrospective self-sorting 4D-MRI scan using a repeated slice-based coronal turbo spin-echo sequence(TR/TE = 316/60 ms, voxel size = 2x2x5 mm3, slices = 25, repetitions = 120, total scan time = 16 minutes). For two volunteers, the stomach was delineated manually for 120 repetitions. For the other two volunteers, the first 10 repetitions were manually delineated. A deep neural network, based on U-net architecture, was trained on these manual delineations. All remaining delineations were generated by this network.

A rigid translation was performed on the contours, of equal size but opposite direction to the liver dome CC position, to correct for respiratory motion. Subsequently, the remaining rigid motion was analysed by calculating: 1) the centre of mass (COM) for all contours, 2) the standard deviation (SD) over the COM per slice and 3) the mean of these COM SDs over all slices of the four volunteers. Non-rigid motion was characterized by: 1) absolute gastric wall position SD in eight equiangular points from the COM (Q1-8) and 2) gastric wall motility, defined as the squared difference between absolute gastric wall position SD and COM SD. Figure 1 shows a coronal MR image along with the gastric contour, Q1-8 and COM.


Results

Figure 2A shows the average motion standard deviations for the four volunteers. For the COM SD, the median ± interquartile range (IQR) over all slices and volunteers was 2.0 ± 0.8 mm and 2.2 ± 0.5 mm in LR and CC directions, respectively. The median gastric wall motility ± IQR was 1.5 ± 0.4 mm and 1.4 ± 0.4 mm in LR and CC directions, respectively. Respiratory CC motion variability was considerably higher than CC wall motility, on average with a factor of 2.8. Gastric motion differed considerably between stomach segments (Figure 2B). The higher LR variability in Q1 and Q5 is mainly due to out-of-plane motion.


Conclusion

This study indicates that rigid motion dominates intra-fractional gastric wall motility and is strongly correlated with respiratory motion. Rigid motion is a factor 2.8 larger than non-rigid motion in the CC direction.