SCAPIS AI platform
There is a significant need for secure access to high-quality medical data and powerful computing resources to allow for the research and development of AI solutions which will lead to more individualised care, improved diagnostics and more predictive measures. SCAPIS is a unique Swedish population study funded by the Swedish Heart-Lung Foundation which has evaluated the condition of the hearts and lungs of 30,000 randomly selected men and women aged between 50 and 64. The SCAPIS AI platform aims to create a secure research environment with access to the relevant infrastructure, tools and processes for AI development, initially on the basis of image data from SCAPIS. The project is led by the Sahlgrenska Academy at the University of Gothenburg. Other partners in the project are Analytical Image Diagnosis Arena (AIDA) at Linköping University and AI Innovation of Sweden.
The SCAPIS database contains a set of data that is the only one of its kind in the world. In Sweden there are several other collections of high-quality medical data where AI can give a better understanding of disease and, importantly, identify possible predictive and preventive measures. The problem is that medical data is sensitive by its very nature. It is often essential to follow up individual patients in the future, which means that data cannot always be anonymised. Therefore, it must be possible to carry out the research without putting personal integrity at risk. In this area we are looking at two different solutions. One is a secure, enclosed environment for AI development and the other involves the generation of synthetic data that corresponds to the SCAPIS images without being actual images. The technology behind deepfake videos will be used to generate new medical images that can be shared more openly.
The objective of the SCAPIS study is to be able to identify individual risks of, for example, stroke, COPD, sudden cardiac arrest, myocardial infarction and other heart diseases. The aim is to acquire much greater knowledge of the origins of the disease in order to be able to prevent it before it occurs. AI solutions can make a significant contribution to achieving this goal.
The project is still in its start-up phase. Its long-term objective is to develop a better understanding of how heart, vascular and lung diseases occur and how they can be prevented. The project will last for two years and, as well as establishing a secure research environment and generating synthetic data, it will run three pilot studies to test the infrastructure and the potential for analysing SCAPIS images.
The first pilot study will develop an automated algorithm for annotating images from a unique investigation carried out as part of SCAPIS involving coronary angiography. This investigation gathered information about the build-up of fatty substances in the coronary arteries. A complete analysis involving manual outlining of all the anatomical and pathological structures in the heart would take around an hour. It would be impossible to do this for 30,000 cases. A fully automated algorithm is therefore essential in order to interpret this data.
The second pilot study relates to epicardial fat. This is a layer of fat that surrounds the heart and its blood vessels. Data is available which indicates that too much fat around the heart accelerates the build-up of fatty substances in the arteries and can cause myocardial infarction. The aim is therefore to segment the fat so that its volume can be measured and linked to indications of coronary artery disease. It takes between one and two hours to manually segment the volume of the fat and therefore fully automated models are needed to process all the images from SCAPIS.
The third image project will be announced by AIDA. It may concern changes in the lungs or the segmentation of the heart pathology. One important aim of the third evaluation project is to allow researchers who did not take part in the development of the platform to validate its function.