| EPSRC Reference: |
EP/X040186/1 |
| Title: |
Turing AI Fellowship: Ultra Sound Multi-Modal Video-based Human-Machine Collaboration |
| Principal Investigator: |
Noble, Professor A |
| Other Investigators: |
|
| Researcher Co-Investigators: |
|
| Project Partners: |
|
| Department: |
Engineering Science |
| Organisation: |
University of Oxford |
| Scheme: |
EPSRC Fellowship - NHFP |
| Starts: |
01 October 2023 |
Ends: |
30 September 2028 |
Value (£): |
4,248,942
|
| EPSRC Research Topic Classifications: |
| Artificial Intelligence |
Human-Computer Interactions |
|
| EPSRC Industrial Sector Classifications: |
|
| Related Grants: |
|
| Panel History: |
|
|
Summary on Grant Application Form |
As artificial intelligence (AI) starts to be deployed in clinical practice, there are big questions about how this will change decision-making in healthcare. AI for full automation of tasks has received most attention, whereas what is arguably more useful in healthcare is for AI to enable humans and machine to share tasks and decision-making. This shifts the view of AI as an automation technology that replaces human skill, to one that empower individuals; with the AI acting as a teacher, peer or assistant depending on context. However, building human-machine collaboration systems for healthcare is hard, in part as the human decision-making in healthcare can be complex, and understanding of how to do this is in its infancy.
A first scientific aim of this fellowship is to develop human-machine collaboration methodology that knows when to defer to a human expert and that can model human-machine shared tasks. To focus the methodology on real-world need I will explore this topic in the context of ultrasound imaging which is a skill that is difficult to master due to the complex hand-eye co-ordinations required to scan well. My group is uniquely placed to conduct this research as we have created a large-scale multi-modal dataset of ultrasound video, and gaze, probe motion and audio data on human operators (sonographers) who performed the scan on a previous project. The research will work towards design of different human-machine collaboration models which interplay the teacher, peer and assistant roles of human and machine. The ultimate goal would be learning-based solutions that can evolve the human-machine collaboration relationship with a user over time.
A second scientific aim is motivated by a practical challenge in data-hungry healthcare learning-based research which is that data governance rules can prevent sharing of data between research sites. A technique called federated learning has recently been proposed to model real-world problems from de-centralised heterogenous data. However, current understanding of how to build robust and principled federated learning models for imaging tasks for the (imbalanced) small data case typical of medical imaging is limited. The second fellowship aim is to develop the scientific foundations of principled FL-based solution design for supervised and unsupervised image and video analysis tasks as well as multi-modal imaging. The ultimate goal is to be able to seamlessly model any imaging problem with decentralised, heterogenous data to the same (or acceptable) accuracy as if the data had been available in one centre while preserving data privacy, or to understand the bounds on when this is possible.
Human skill, and differentiation of human skill is poorly understood. How to measure human skill at a task in an objective way is also a largely open research question. The third fellowship aim is to develop video and multi-modal learning-based tools to objectively study human skill at performing healthcare tasks. This research will bring together researchers with complementary interests in wearable sensors, computer vision, medical image analysis, clinical medicine (ultrasound imaging and surgery) and industry (training simulation and wearable sensors) to advance the state-of-the-art and will explore as well implications on how AI adoption is changing expectations on human skills.
Beyond the science created, the fellowship will contribute to AI scientist training. The human skill modelling workstream includes a Healthcare Research AI Ambassador fellowship scheme for researchers with a clinical or humanities (AI ethics) background to work with AI scientists to develop human skill models for clinical tasks and ethics of AI related to human-machine collaboration. An industry-academia mobility scheme is also proposed that aims to address retention of talent in academia, early-career researcher mobility, and encourage greater interaction between UK academia and industry in healthcare AI.
|
|
Key Findings |
|
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
|
|
Potential use in non-academic contexts |
|
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
|
|
Impacts |
|
| Description |
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk |
| Summary |
|
| Date Materialised |
|
|
|
|
Sectors submitted by the Researcher |
|
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
|
| Project URL: |
|
| Further Information: |
|
| Organisation Website: |
http://www.ox.ac.uk |