Innovative data science solutions to improve cancer care and outcomes in Scotland Phase 2 Project
Canon Medical Research Europe has been awarded further funding of £140k to undertake a 6 month development and evaluation of a proof of concept/prototype starting from 23 July 2018
Canon Medical Research Europe
Automatic RECIST reporting in Mesothelioma using Deep Learning AI
RECIST (Response Evaluation Criteria in Solid Tumours) is a scoring system applied to CT scans to describe a patient’s response to cancer treatment, and is the gold standard measurement used in clinical trials. However, RECIST is time-consuming and results often vary between reporting radiologists. A shortage of NHS radiologists also means RECIST is not routinely used in NHS care. Our aim is develop an automatic RECIST using Artificial Intelligence (AI), improving the quality and reducing the cost of cancer response assessment. The AI constructed will use deep convolutional neural networks (CNNs), after careful training using human inputs.
Malignant Pleural Mesothelioma (MPM) is an asbestos-related cancer with a high incidence in the UK. MPM develops in the lining of the lungs (the pleura), and exhibits an unusual, rind-like growth pattern. Human RECIST reporting is particularly challenging in MPM and a more robust RECIST method would be a major advance in this disease.
Working closely with our collaborators at NHS Greater Glasgow and Clyde’s Department of Respiratory Medicine, Canon Medical will use their decades of experience in medical image analysis to develop state-of-the art tumour segmentation methods for CT, specifically targeted to support RECIST reporting for MPM. CT images from two recent multi-centre studies (DIAPHRAGM and SWAMP) will be used to train Canon Medical’s novel AI in tumour segmentation, using paired MRI scans for high quality reference information. The accuracy of the new AI be tested by blinded validation, using 380 CT scans pre- and post-chemotherapy from the PRISM study, in which human RECIST scores are recorded in triplicate, providing a robust human reference standard.