ESTRO meets Asia 2024 Congress Report
The physics track at ESTRO meets Asia started with a discussion on the current challenges that are associated with the use of dose mapping and accumulation in clinical practice led Dr Eliana Vasquez Osorio.
Dose mapping refers to the process of transferring dose distributions from one image to another, which is done using deformable image registration (DIR). Dose accumulation is the next step, during which the mapped dose is summed, voxel by voxel, with a second dose cube to produce an estimation of a total dose distribution. Dose mapping/accumulation is highly desirable for multiple applications in radiotherapy, for example, to estimate delivered doses in fractionated treatments and replan scenarios, and to assess cumulative doses in treatment combinations and in the reirradiation setting.
DIR and dose mapping
Since DIR technology has been in clinical software for several years, the question is why dose mapping/accumulation is not widely used. One major challenge is DIR uncertainties. These uncertainties directly impact dose mapping, as registrations that produce slightly different results that still ‘optimally’ align the images can produce (slightly) different mapped dose distributions. In many cases, there is no way to determine which mapped dose distribution is better, as there is no ‘ground truth’ to compare against.
DIR uncertainties
DIR uncertainties can be caused by large anatomical changes and image acquisition artefacts, but also by the choice of the initialisation, which is commonly done using rigid image registrations (RIR). Sub-optimal initialisation is one reason for DIR failure. The key insight is to remember that RIR initialisations are used to ‘ease’ the job of a subsequent DIR. As RIR results tend to be optimal for only small parts of the anatomy, you need deliberately to choose to focus the RIR initialisation on the region of importance rather than simply keeping the default values. Interestingly, even when the initialisation plays such an important role, it is ignored or under-reported in the current literature.
Uncertainties also come from the assumptions that are made during DIR implementation (in particular continuity and smoothness). It is assumed that the changes in each anatomical location are very similar to changes in locations in the vicinity. This assumption is broken for any sliding interface, e.g., between the lungs and thoracic wall or among abdominal or pelvic organs. Due to this limitation, the dose that is mapped for neighbouring organs that slide across each other is likely wrong and must be taken with extra care. A second assumption is that the same tissue is present in both images; again, this assumption is often broken in real data, when external devices are introduced (e.g., brachytherapy applicators) or after other interventions (e.g., surgery). Advanced registrations have been proposed in the past to deal with these non-continuous, non-smooth changes and for tissue mismatch, but unless these are available in commercial software, their potential remains untapped.
Other points that were raised during the lecture were: special considerations for adaptive radiotherapy (changes in the tumour that are sometimes not visible in the images, and the registration strategies), parameters as a source of DIR uncertainties, and the role of contour consistency for both assessment and/or guidance for registrations.
Dose gradients enhance the impact of uncertainties
During the lecture and through examples, Dr Vasquez Osorio showed how the areas more prone to dose mapping uncertainties are those where dose gradients are present. Different approaches to account for them have been proposed. These come down to estimating the worst-case value for each voxel (either by selecting the local maximum dose within a sphere centred on each voxel of organs at risk or based on standard deviations from repeated registrations). Similar to the current situation regarding advanced registration methods, unless these tools are accessible via commercial software, clinical users will not be able to benefit from them.
Beyond DIR uncertainties, and more relevant to dose accumulation, is the need to correct for differences in fractionation. This is critical in applications such as reirradiation. Multidisciplinary research is needed here to standardise this step.
Clear communication
A final message is that communication is vital. Clear specification of the intended use and accuracy requirements, how to quantify these measures, summarise them and report them is important. Also, a clear report on the uncertainties both for DIR and dose mapping is necessary for clinicians to make informed decisions on whether or not to use the mapped dose. For this, a specific request/report form, such as one that extends those previously proposed in the TG-132 report, is required.
The talk was followed by a lively discussion on aspects related to automation, how to best document different aspects, the best way to flag sub-optimal registration results, and what further efforts are required to make this more accessible to the clinical user. All these issues are avenues for future clinical and academic research and implementation.
So… are we there yet?
Well… not really. Dose mapping/accumulation methods with tools to quantify uncertainties are urgently needed for many clinical applications. However, to have meaningful clinical impact, reliable tools need to be developed and made accessible from commercial software.
Further reading:
Nenoff L, Amstutz F, Murr M, Archibald-Heeren B, Fusella M, Hussein M, Lechner W, Zhang Y, Sharp G, Vasquez Osorio E. Review and recommendations on deformable image registration uncertainties for radiotherapy applications. Phys Med Biol. 2023 Dec 13;68(24):24TR01. doi: 10.1088/1361-6560/ad0d8a. PMID: 37972540; PMCID: PMC10725576.
Murr M, Brock KK, Fusella M, Hardcastle N, Hussein M, Jameson MG, Wahlstedt I, Yuen J, McClelland JR, Vasquez Osorio E. Applicability and usage of dose mapping/accumulation in radiotherapy. Radiother Oncol. 2023 May;182:109527. doi: 10.1016/j.radonc.2023.109527. Epub 2023 Feb 10. PMID: 36773825.
Lowther N, Louwe R, Yuen J, Hardcastle N, Yeo A, Jameson M; Medical Image and Registration Special Interest Group (MIRSIG) of the ACPSEM. MIRSIG position paper: the use of image registration and fusion algorithms in radiotherapy. Phys Eng Sci Med. 2022 Jun;45(2):421-428. doi: 10.1007/s13246-022-01125-3. Epub 2022 May 6. PMID: 35522369; PMCID: PMC9239966.
Dr Eliana Maria Vasquez Osorio
Division of Cancer Sciences, Faculty of Biology, Medicine and Health
The University of Manchester
Manchester, UK
email: eliana.vasquezosorio @ manchester.ac.uk
LinkedIn profile: https://linkedin.com/in/elivaos
X handle: @elivaos