The lack of an effective early diagnosis tool and the challenge of staging (i.e. measuring the size) of Malignant Pleural Mesothelioma (MPM) prompted a proposal from Canon Medical Research Europe and the Glasgow Pleural Disease Unit to the Scottish Cancer Innovation Challenge in mid-2018.
Using phase one funding, the study team tested the feasibility of a method to automatically identify MPM tumours and their boundaries on CT images (a process known as ‘image segmentation’). Subsequent phase two funding has allowed development of a prototype algorithm that uses a trained Artificial Intelligence (AI) system to achieve this. This development paves the way for an AI system that could greatly improve the efficiency of clinical trials and the accuracy and reliability of measurements of response to treatments in the clinic.
Dr Kevin Blyth, Honorary Clinical Associate Professor at the University of Glasgow and Chief Investigator of the study, said:
“We are collaborating with Canon Medical Research Europe, who have considerable expertise in use of AI for medical imaging via the Cancer Innovation Challenge. This fantastic initiative facilitates closer integration between industry partners and clinical researchers.
“Automatic RECIST (Response Evaluation Criteria in Solid Tumours) reporting is a scoring system applied to CT scans to describe a patient’s response to cancer treatment but is difficult to use in mesothelioma, because of the complex shape of mesothelioma tumours. A carefully trained, and properly optimised AI algorithm could theoretically locate a mesothelioma on a CT scan and measure its volume, allowing comparison with previous measurements and making clinical trials less expensive. Development of this system involves a training process in which a human must ‘draw around’ all areas of mesothelioma on each CT scan to show the AI what this looks like. In phase one we learnt this process (known as ‘ground truth generation’) needed to be done by an expert mesothelioma clinician, with experience of mesothelioma images and the anatomy of the chest, since it proved very difficult for an imaging technician.
“This training and AI development process is going well, and we hope to be able to announce our results in the autumn of this year. If we are successful at this stage, we will be ready to test our trained AI algorithm in larger numbers of CT scans, bring us closer to deployment of AI RECIST in future trials and clinical practice.”
For more information on the Cancer Innovation Challenge visit www.cancerchallengescotland.com