A unique research study within medical image analysis of coronary/cardiac computed tomography angiography images from the Scapis study has been tested this autumn at AI Innovation of Sweden's Data Factory. By using the infrastructure that the Data Factory offers, it has been possible to dramatically increase efficiency in the research work.
Within SCAPIS, data from about 30,000 people has been collected at six university hospitals in Sweden, wherein each study participant was mapped with different samples and examinations and also had to answer a variety of questions in surveys. The analyses initiated by the material should provide answers to why diseases such as stroke, COPD, sudden cardiac arrest and myocardinal infarction occur, why some recover from them and others do not, and how to prevent them from occurring.
The SCAPIS study is unique in that it includes data from a series of physiological examinations, x-ray images, blood samples and responses from questionnaires from a comprehensive number of individuals. The large amount of data, a significant amount of which includes images, requires automatic methods for analysis that can hopefully be both faster and more consistent than manual evaluators.
By using a technical infrastructure like that of AI Innovation of Sweden's data factory, the training of deep learning algorithms can be made drastically more efficient as an important component in automatic analysis.
Jennifer Alvén, doctoral candidate at Chalmers within the field of medical image analysis, is working on a project in which she develops automatic algorithms for analysis of SCAPIS images. In this particular case, it involves using deep learning in order to automatically outline coronary arteries in CTA (computed tomography angiography). Part of the training of the deep learning algorithms has now been carried out on the GPU cluster in AI Innovation of Sweden's data factory.
"I hope that we will be able to present a method that is at least as good as a manual evaluator at outlining coronary arteries and detecting and classifying stenosis (narrowing of the coronary arteries) and plaques," Jennifer explains.
With access to the data factory, it has gone significantly faster training my deep learning algorithms - from about 24 hours to 1 hour per training. It also frees up my regular computer so that I can use it for other purposes.
"Practically speaking, it is not possible to manually analyse the large amount of data. For example, it would take up to a couple of hours for an radiologist/radiographer to outline coronary arteries for an individual (i.e. one image)."
Jennifer's research project within SCAPIS is the largest that has been implemented in the Data Factory's infrastructure. The work has been in progress this autumn.
Johanna Bergman, project manager at AI Innovation of Sweden, is very happy that SCAPIS chose to carry out their part of their work here.
"The work Jennifer is doing is important for medical research and for SCAPIS, and it is also a concrete example of how machine learning has potential within medical image analysis.
To now have the opportunity to test the infrastructure at The Data Factory has meant a major step forward for us. Because the SCAPIS project involves handling medical and personal data, there are very big demands on the project, not only in terms of speed, but also with respect to security and the legal aspects.
So, for our part, it has been an affirmation that we meet the requirements posed by medical image analysis and medical research."