| EPSRC Reference: |
EP/Y018036/1 |
| Title: |
People-centered Mammogram Analysis |
| Principal Investigator: |
Carneiro, Professor G |
| Other Investigators: |
|
| Researcher Co-Investigators: |
|
| Project Partners: |
|
| Department: |
Vision Speech and Signal Proc CVSSP |
| Organisation: |
University of Surrey |
| Scheme: |
Standard Research - NR1 |
| Starts: |
02 October 2023 |
Ends: |
01 April 2025 |
Value (£): |
435,211
|
| EPSRC Research Topic Classifications: |
| Artificial Intelligence |
Medical Imaging |
|
| EPSRC Industrial Sector Classifications: |
|
| Related Grants: |
|
| Panel History: |
|
|
Summary on Grant Application Form |
The use of artificial intelligence (AI) to automate routine medical image analysis tasks, such as the analysis of mammograms, has been proposed to address the Radiology crisis because of low recruitment rates, the 10-year training time to produce a consultant radiologist, and the pending retirement of large parts of the current workforce. The design of these AI models tends to be centred on data rather than people, i.e., the key stakeholders patients and radiologists, where model optimisation focuses on maximising the accuracy of a generic model for the majority of patients instead of improving the performance of all sub-groups of radiologists for all patient cohorts. Note that radiologists can be divided into sub-groups depending on their experience and track record, while patients can be grouped into cohorts subject to their age, family history, previous cancer diagnosis, ethnicity, and scanner characteristics. Such design strategy has produced accurate models for the majority of patients, but its intrinsic competitive nature with radiologists' performance and lack of consideration for particular sub-groups of radiologists and patients cohorts have prevented the widespread adoption of AI models in clinical practice.
We argue that the two main reasons for this weak acceptance of AI into clinical practice are: 1) radiologists have not been integrated in the design and optimisation of the models, resulting in poor cooperation between radiologists and models; and 2) patients are provided with a biased classification performance since the system performs well for the majority of patients. This proposal addresses these two problems, targeting a more usable AI model that will increase the sensitivity and specificity for all sub-groups of radiologists, improving the efficiency of the whole mammogram analysis process, and potentially allowing radiologists to join the clinical force earlier in their careers to mitigate the Radiology crisis. Also, given that patients form cohorts in unpredictable ways (depending on dataset population and imaging technology), patients will be more likely to accept AI models that produce a fair outcome for all cohorts. We focus this project on mammography, but our proposal is applicable to other similar problems, such as lung cancer screening. We will work with our collaborators from NHS and industry on an extension of our method to these similar problems during the development of this project.
|
|
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.surrey.ac.uk |