Why did we fund this project?
This award aims to implement an autosegmentation tool in magnetic resonance imaging (MRI) to reduce the number of KPC mice, the most common pancreatic cancer mouse model, needed in orthotopic pancreatic cancer studies.
Research on pancreatic cancer and the development of treatments predominantly uses mice, including genetically altered animals where tumours spontaneously develop. Longitudinal studies are performed to analyse the tumour, which typically requires mice to be culled at each time point. The use of MRI would allow tumour development to be studied in the same animals, however, the interpretation of images can be challenging as the mouse pancreas is soft and diffuse making it difficult to define from surrounding tissues. Through an NC3Rs PhD Studentship, awarded to Professor Jane Sosabowski (Co-Investigator) Algernon Bloom has used machine learning to develop an autosegmentation tool (3D CAMMP) that can automatically identify a tumour-positive pancreas in an MRI image with 95% accuracy compared to veterinary radiologists and image analysis experts, allowing easier analysis of MRI imaging. Mice can be entered into a study based on tumour size (rather than animal age) enabling a more accurate tumour assessment and a subsequent reduction in tumour variability. Using MRI imaging, the Barts Cancer Institute has reduced the number of KPC mice per pancreatic cancer study from 12 to eight animals.
Algernon will now collaborate with researchers at the Francis Crick Institute, the Beatson Institute of Cancer (University of Glasgow), and the University of Cambridge to implement 3D CAMMP in their laboratories. Algernon will also develop a user-friendly interface to enable uptake by non-imaging experts, in collaboration with Invicro, a Boston-based image analysis and software development company. 3D CAMMP is available to users through the Invicro Whole Body Atlas.
This award was made in collaboration with Cancer Research UK.
Pancreatic cancer is extremely difficult to treat with only about 5 % of patients surviving for 5 years. New treatments are urgently needed and many cancer researchers are developing and testing new therapies. These researchers often use specially bred mice to study pancreatic cancer. These mice are born with a healthy pancreas but develop pancreatic tumours deep within their body as they age. Its possible to detect these pancreatic tumours from outside the body using imaging and this can be done so it has very little effect on the animal. Imaging allows scientists to study the tumour growth and whether or not it responds to new therapies.
However, pancreatic tumours can be extremely hard to detect for scientists who are not experts in imaging. Therefore we have developed an artificial intelligence computer program that can find pancreatic tumours on MRI scans automatically with very little input from the researcher. This allows for faster and more accurate tumour detection. Because the tumour can be measured more accurately we can reduce the numbers of animals used in the study and detect tumours earlier when they have less impact on the animal's health.
The small MRI instrument that was used to develop the computer program allows scientist to perform the scans themselves after minimal training. In this project, we are transferring the previously developed method to accurately measure the tumours to three other cancer institutes in the UK. We think that using the artificial intelligence program will allow them also to improve the accuracy of their studies while also reducing the numbers of animals they use in pancreatic cancer research.
InstitutionQueen Mary University of London
Co-InvestigatorProfessor Kairbaan Hodivala-Dilke
Dr Jane Sosabowski