GuidelinesAn evaluation of MR based deep learning auto-contouring for planning head and neck radiotherapy
Section snippets
Methods
Bilateral parotid glands and submandibular glands (SMGs) were contoured by the same clinician on one hundred anonymised oropharynx T2 weighted (T2W) 2 dimensional (2D) diagnostic MR scans. Diagnostic T2W MR scans were used due to the ease of obtaining sufficient numbers to investigate the model’s performance. To ensure adequate visualisation of structures, oropharynx scans were excluded if the images contained artefacts or there was tumour involvement of the parotid or SMGs. Contours were peer
Comparison of manual contours and modelMRI
ModelMRI automated contours showed good agreement with manual contours defined on T2W 2D diagnostic and T2W Dixon 2D radiotherapy planning scans for bilateral parotid glands and SMGs (mean DSC ≥ 0.80). The agreement was lower for the T2W 3D MRL for the bilateral parotid glands (mean DSC 0.70). There was a lack of overlap between modelMRI automated and manual contours on the MRL images for the left and right SMG (mean DSC of 0.10 and 0) as shown in Supplementary Table 2. The overlap of manual
Comparison of modelMRI and modelCT
ModelMRI contours showed better agreement with manual contours on MR radiotherapy planning scans compared with modelCT contours on CT radiotherapy planning scans. An improvement in mean DTA and reduction in SD for bilateral parotid and SMGs was observed except for the left SMG which was similar between the two modalities (see Fig. 4). The mean DSC values to compare the model and manual contours were similar for the bilateral parotid and left SMGs irrespective of the model used (mean DSC > 0.7).
Discussion
This study is the first to evaluate the cross-acquisition of a deep learning MR auto-contouring model on multiple MR sequences for head and neck OAR auto contouring. The work shows that MR based deep learning auto-contouring models show promise as an aid to clinician OAR contouring but are sensitive to the MR sequence used. ModelMRI performed well when applied to T2W 2D diagnostic and radiotherapy planning scans, with a mean DSC score of 0.80, and DTA of ≤ 2 mm.
The MR auto contouring model was
Conclusion
MR based deep learning auto-contouring models show promise as an aid to clinician OAR contouring. A model trained on diagnostic MR images has been shown to work well on RTP images as well as diagnostic images. However extending this to MRL images shows that these models remain sensitive to the MR sequence used. Further work is needed to optimise the MRL image acquisitions and evaluate the model in a larger cohort prior to evaluating its use in adaptive re-planning.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
The Christie NHS Foundation Trust is supported by a Cancer Research UK Centres Network Accelerator Award Grant (A21993) to the ART-NET consortium. Professor West and Professor van Herk are supported by the NIHR Manchester Biomedical Research Centre. This work was also supported by Cancer Research UK via funding to the Cancer Research Manchester Centre [C147/A25254].
The MR-Linac images were taken as part of the PRIMER imaging protocol funded by RMH/ICR NIHR biomedical research centre.
References (24)
- et al.
Improving automatic delineation for head and neck organs at risk by Deep Learning Contouring
Radiother Oncol
(2020) - et al.
Does atlas-based autosegmentation of neck levels require subsequent manual contour editing to avoid risk of severe target underdosage? A dosimetric analysis
Radiother Oncol
(2011) - et al.
Prospective randomized double-blind study of atlas-based organ-at-risk autosegmentation-assisted radiation planning in head and neck cancer
Radiother Oncol
(2014) - et al.
Deep learning-based delineation of head and neck organs at risk: geometric and dosimetric evaluation
Int J Radiat Oncol Biol Phys
(2019) - et al.
Deep learning algorithm for auto-delineation of high-risk oropharyngeal clinical target volumes with built-in dice similarity coefficient parameter optimization function
Int J Radiat Oncol Biol Phys
(2018) - et al.
Validation of a magnetic resonance imaging-based auto-contouring software tool for gross tumour delineation in head and neck cancer radiotherapy planning
Clin Oncol
(2017) Head and neck MR imaging
Neuroimag Clin N Am
(2004)- et al.
Evaluation of head and neck tumors with functional MR imaging
Magn Reson Imaging Clin N Am
(2016) - et al.
Adaptive radiotherapy: The Elekta Unity MR-linac concept
Clin Transl Radiat Oncol
(2019) - et al.
Clinical validation of atlas-based auto-segmentation of multiple target volumes and normal tissue (swallowing/mastication) structures in the head and neck
Int J Radiat Oncol Biol Phys
(2011)