Q Exchange
AI Auto-Contouring for Radiotherapy
- Shortlisted idea
- 2022
Meet the team
Also:
- All North West Anglia NHS Foundation Trust staff:
- • Jamie Fairfoul, Head of Radiotherapy Physics
- • Mark Cowen, Head of Radiotherapy Treatment Planning
- • Andy Mills, Deputy Head of IT (Projects)
- • Elpiniki Papadimitriou, Senior IT Project Manager,
- • Katie Jephcott, Clinical Lead for Oncology
- - David Parker-Radford, Head of Quality Improvement
What is the challenge your project is going to address and how does it connect to the theme?
All radiotherapy patients require target volumes and organs at risk from irradiation to be outlined by consultant oncologists. This can be an extremely time-consuming process, taking up to three hours for a complex case, and can be the source of variations in quality. With increasing patient numbers, increasing complexity of cases, and national long-term staffing pressures on the oncology workforce, contouring is at risk of becoming an activity-limiting step in our pathways. We would like to implement an AI auto-contouring solution, which would outline structures automatically, leaving oncologists to quality assure the outlines and adjust where clinically appropriate. This would generate a significant time and cost saving (evidence suggests up to 80% time saved per patients), whilst simultaneously increasing quality and consistency in outlining across our patient group.
What does your project aim to achieve?
This project aims to reduce time taken to complete radiotherapy outlining of complex sites by using an artificial intelligence solution, starting with patients on the colorectal and breast pathways at North West Anglia NHS Foundation Trust.
The objectives are:
– To implement and test an artificial intelligence solution for contouring targets and organs at risk in over 500 radiotherapy patients.
– To reduce the radiotherapy treatment pathway from 31 days to 17, for category 1 patients.
– To reduce oncologist and dosimetrist task time spent on contouring, freeing up time for additional patient capacity and more complex treatment planning.
– To share learning with regional and national networks, using quality improvement templates
This project aims to also reduce variation and improve consistency of approach to contouring and ensure patient safety through human validation at every step, peer review and a thorough evaluation to establish the impact of machine intervention on patient outcomes.
How will the project be delivered?
The project consists of six phases:
– Phase 1: Preparation including stakeholder planning and engagement communications (3-4 weeks)
– Phase 2: Technical Commissioning and Clinical Validation (up to 2 months); In this phase the solution is commissioned, with the technical requirements to be embedded successfully within existing systems safely and compatibly. Clinical validation of output will build user confidence and set limits on its application.
– Phase 3: Timing Studies (2 months) – This is the first live tests of the capabilities of the AI in managing the required workload in a timeframe that improves speed in a high quality way.
– Phase 4: Impact Measurement (2 months) – This consists of 1 month data collection and 1 month validation
– Phases 5 and 6: Share and Embed Learning (6 months) – Project streamlining to business as usual; communications to networks. Based on data, managed rollout to other tumour sites
How is your project going to share learning?
The project plans to share learning with the wider Q community through blog posts. As a new implementation of this technology, it will share learning among a regional network of radiotherapy centres, and also at national oncology courses and learning events. As Quality Improvement work, it will be showcased on the internal Quality Improvement Academy at the Trust through a poster and audio storytelling of the improvement journey. The project seeks to gather intelligence from consultant oncologists and dosimetrists in particular on the role this technology has played in improving workplace wellbeing.
How you can contribute
- Sharing their experience of testing AI solutions
- Sharing experience of quality control/assurance of AI solutions in radiotherapy
- Relevant expertise
Plan timeline
| 24 Jun 2022 | Phase 1 Start: Preparation including stakeholder planning and engagement communications |
|---|---|
| 24 Jun 2022 | Receive Q Funding |
| 22 Jul 2022 | Phase 2 Start: Commission solution and technical requirements |
| 22 Sep 2022 | Phase 3 start: Timing Studies to test AI capability in action |
| 22 Nov 2022 | Phase 4 start: Impact measurement and validation |
| 5 Feb 2023 | Phase 5-6 start: Share and embed learning. Test rollout/safe scaling up |
Comments
Shereen Nabhani Gebara 23 May 2022
Good luck with this submission
Using AI to support and streamline cancer care is very timely and important.
I am currently working on an EU funded project (https://incisive-project.eu/) for using AI for the early detection of cancer.
So much potential
David Parker-Radford 23 May 2022
Hi Shereen,
Thanks very much for sending through this link. We agree there is definite potential in terms of processing speed and better patient outcomes providing there is a planned, and safe human-led rollout of the technologies with human clinical review a key safeguard.
This works sounds really interesting and I have forwarded onto our fabulous Radiotherapy team, in relation to this project.
Kind thanks
David
Helen Meese 17 May 2022
hi David, I dont know if this is of interest, or relevant to your work, but I was at a meeting today with Dr James Squires, Head of Policy at the Academy of medical Sciences and he is undertaking some policy research to investigate the Impact of AI on Research & Healthcare.
I think he would be very interested to learn more about your work. If you want to get in touch with him he can be contacted at james.squires@acmedsci.ac.uk
If you mention Dr Helen Meese from the Royal Academy of Engineering's Healthcare Policy Panel suggested you get in touch, I think he would be very pleased to hear from you.
David Parker-Radford 17 May 2022
Hi Helen,
Thanks for letting us know. It would be interesting to hear more about this research. Thank you very much for building this connection. His email address is publicly available so I'll speak to my project team and we may reach out to him (if we are successful with the funding) to explore the potential for collaboration further.
Kind thanks
David
David Parker-Radford 13 May 2022
As a project team, we are very excited by the potential of our project and we are excited that we have reached the shortlist stage. We are all lined up and ready to start the project as soon as the funding is secured.
David Parker-Radford 13 May 2022
If you like this project and want to see us funded please do share the link to this page with other members of your networks.
David Parker-Radford 5 May 2022
Thanks for the question and I hope we helped answer your question Sarah. Do feel free to post again or get in touch if you have any further questions.
David Parker-Radford 5 May 2022
The training datasets are generated separately for different body regions, e.g. male/female pelvis, thorax, head & neck, breast + nodes, etc. The manufacturer works in collaboration with clinical experts, creating library of test cases for each body site, working to international consensus guidelines for organ outlining so that the AI models are trained on current best clinical practice. The AI models are constantly in development, to include, for example, different imaging modalities or adjustments to clinical best practice based on new evidence.
The AI will have application and benefits across a range of case complexities – simple cases will need minimal intervention after auto-contouring, whereas more complex cases such as head and neck or breast and nodes will need a greater degree of verification and potential adjustment by the specialists, but the net time saving will be significant and will enable increased patient throughput and more consistent results.
Sarah Mackie 4 May 2022
Important project and seems to have obvious value. What kind of training dataset did/does the AI use and does it use similar machines to those you use locally? Is the AI potentially suitable for all cases including the most complex or is the idea to use it to speed up the simpler cases, freeing up time for the specialists to work on contouring the really tricky/complex cases?
Comments are now closed for this post.