Innovative data science solutions to improve cancer care and outcomes in Scotland
Three projects were awarded funding of up to £35k each to undertake a 3 month feasibility study from 5 March 2018. At the end of Phase 1, these projects will be invited to apply for Phase 2 funding where 1 or 2 projects will be awarded funding of up to £125k each to develop, evaluate a prototype of their project over a 6 month period from July 2018.
Canon Medical Research Europe Ltd
Automatic RECIST measurement in mesothelioma by deep learning
‘RECIST’ is a widely used scoring system for assessing response to cancer treatment, and is considered the gold standard in clinical trials of new therapies. However, it is time-consuming and is not always conducted as part of routine NHS care, because of limited radiologist reporting time. RECIST is also subject to inter-reporter variability, which can impact on treatment decision and trial outcomes. Our project aims to enable the introduction of automated RECIST reporting within the NHS, thus improving patient care and providing quantitative measures and analysis of patient outcomes.
Malignant Pleural Mesothelioma (MPM) is an asbestos-related cancer with particularly high incidence in Scotland. The lack of a robust response assessment tool is a major limitation in evaluating new therapies for mesothelioma and an accurate, efficient method is urgently required.
We will develop a system for the automatic measurement of RECIST score in MPM, based on CT imaging and machine learning (deep learning) techniques. We will leverage data and ground-truth (i.e. manually-determined RECIST) from the PRiSM (Prediction of Resistance to Chemotherapy using Somatic Copy Number Variation in Mesothelioma) study, which has been funded by the British Lung Foundation and is currently selecting previously treated patients in Glasgow and other UK centres.
Our ultimate goal is for a fully automated reporting system. However, in Phase 1 of the project we propose to develop a semi-automated system, requiring user-provided seed points. The system would then automatically perform tumour delineation, measurement and baseline/follow-up comparison. MPM is a particularly difficult test case, thus presenting the opportunity for a robust proof of principle, from which we can expand this work into other cancers and other modalities including Magnetic Resonance Imaging (MRI).
Jayex Technology Ltd
Real-time Cancer Data Access - A Milestone for Precision Medicine Delivery in Scotland
Jayex engaged with NHS National Services Scotland and NHS Lothian stakeholders to work on a project to design and build a Clinical Access Platform (CAP) able to integrate with national data infrastructure and Scottish Cancer Registry. It will complement its functionality through real-time cancer data access for clinicians and policymakers.
Our proof of concept will focus on the Haematology Cancers, as there is a shortfall of this data in the National Registry. We will be standardising and migrating the existing data collected by the clinicians locally over 30+ years - from the outdated, unsupportable system, to the new, cutting edge platform with an innovative structure, mapped to a global data standard.
Working with expert partners Pulse Infoframe Inc., we will include modern technology advances to develop a robust and scalable solution assisting in day-to-day clinical management, policymaking, as well as supporting clinical trial recruitment and clinical research. Integrated advanced analytics tools will enable meaningful data discovery for clinical decision support. With a structure allowing future mapping of genomics and analysis of unstructured data our Platform will also enable adoption of precision medicine approaches.
The Platform will be interoperable and scalable, meaning that the investment can be leveraged across all other cancer types, rare diseases and other data types in the future, to eventually deliver a real difference to the lives of cancer patients in Scotland and beyond.
Sharpe Analytics Ltd
Using Machine Learning to Estimate Outcomes in Scottish Cancer Patients
Machine learning is driving revolution across a great number of fields by unlocking the predictive power of large datasets. Within healthcare, critical clinical decisions rely on analysis and interpretation of various data types, including weight measurements, blood test results, and radiological and pathological findings.
Decisions on cancer treatments in particular require firm evidence and due deliberation, as the considerable side-effects of various therapies have the potential to bring more harm than good to patients. Therefore, a tool enabling clinicians to use data-backed evidence to inform treatment plans would aid patient management. In addition, more accurate predictions on areas of clinical need within cancer services may help guide resource allocation and planning.
Sharpe Analytics will harness the power of machine learning in order to generate tools for the prediction of outcomes for Scottish cancer patients.
In Phase 1 of the Cancer Innovation Challenge, we will begin with prognosis modelling for patients with renal cell carcinoma, using routinely collected data recorded in repositories such as the Scottish Cancer Registry. Secondary prediction models for other outcomes may also be constructed, as guided by the available data.
These efforts will set the foundations for further work to increase the accuracy of our models by incorporating additional variables, such as genetic markers influencing the likelihood of tumour development. We ultimately aim to build similar prediction tools for other tumour types, and for application to patients both within Scotland, and beyond.