Probabilistic lung tumour target definition from 4DCT data: A motion model based approach
Eiben, B. Chandy, E. Abravan, Azadeh Rompokos, V. Grimes, H. D'Souza, D. Poynter, A. McClelland, J. R.
Eiben, B.
Chandy, E.
Abravan, Azadeh
Rompokos, V.
Grimes, H.
D'Souza, D.
Poynter, A.
McClelland, J. R.
Citations
Altmetric:
Abstract
Purpose or Objective
Respiratory motion is one of the largest sources of uncertainty in RT for thoracic and upper abdominal
tumours. 4DCT is commonly used to define an ITV which encompasses the GTV’s shape and motion, however,
this modality is prone to artefacts due to irregular breathing. Furthermore, GTV positions as seen in the sorted
4DCT images only represent its position for one or two breath cycles, i.e. when the tumour is in the field of
view (FOV) of the scanner detector making it susceptible to outliers resulting in potential geometrical miss.
Instead, we propose to use a motion model based target definition. We utilise our unified motion modelling
and image registration framework that – instead of generating a fixed number of phase (or amplitude) images –
uses the unsorted 4dCT data to generate a single, motion-compensated-reconstruction (MCR) image and a
motion model that describes the subject’s internal motion as function of their breathing trace. Hence, a GTV
outlined on the MCR can be animated over the entire 4DCT acquisition using the model, from which a probabilistic ITV can be calculated. This makes the target definition less susceptible to outlier breath cycles
and irregularities.
Materials and Methods
4DCT data from three lung-cancer patients was unsorted according to each slice's time stamp and aligned with
the breathing trace, then used to fit a motion model and generate an MCR. An ITV was contoured on the 4DCT
data (phase images and maximum-intensity projection) following standard clinical practice (ITV-4DCT), and a
corresponding GTV on the MCR (GTV-MCR). The GTV-MCR was transformed by the motion model to predict its
position for every time point of the breathing trace. For each voxel, the probability that it belonged to the
GTV-MCR was calculated, and this was used to form probabilistic ITVs encompassing all voxels with a
probability >10%, >5%, >1%, and >0% (ITV-10%, ITV-5%, ITV-1%, ITV-0%).
Results
The MCR image quality is visually superior to the standard 4DCT phase images in terms of sharpness and shows
no sorting artefacts (fig 1). Fig 2 shows a coronal slice and a close-up of the model generated GTV-MCR
probability map and the contour of ITV-4DCT for all three patients. For patient 1 and 2 the ITV-4DCT is smaller
than all probabilistic ITVs, and for patient 3 ITV-4DCT is similar in size to ITV-5%, (39.7 and 40.0ccm
respectively) but is a different shape to ITV-5% Conclusion
We have demonstrated the feasibility of generating probabilistic ITVs from a motion model built on unsorted
4DCT data and the subject’s breathing trace. Our initial results indicate that the widely used ITV-4DCT may not well represent the actual GTV motion, and that the model generated probabilistic ITVs provide a flexible
solution for defining more robust and suitable motion encompassing targets.
Description
Date
2021
Publisher
Collections
Keywords
Type
Meetings and Proceedings
Citation
Eiben B, Chandy E, Abravan A, Rompokos V, Grimes H, D'Souza D, et al. Probabilistic lung tumour target definition from 4DCT data: A motion model based approach. Radiotherapy and Oncology. 2021;161:S731-S3.