In this hackathon, we build upon recent advancements in the field of automated segmentation on the 4D (spatio-temporal) MRI data. This work is currently performed in collaboration with the cardiovascular department of the University Medical Center Groningen.
Pim van der Harst, professor of Intervention and translational cardiology, “We are on the threshold of data-driven health care. You see more and more specific applications for the organization, diagnostics or treatment of patients.”
“Even though a hospital with a cohesive AI system that optimizes the organization and care doesn’t exist yet, it’s no longer to question if it will happen, but who will deliver it and when. By organizing a hackathon during DDSW, we hope to speed up this process.”
To alleviate the medical staff from having to manually paint the regions of the left- and right- heart chambers and the heart muscle. The deformation of the heart regions during a heartbeat is indicative of deciding for a pacemaker or assessing the outcome of medicine usage. Automating the image segmentation task and the extraction of characteristic metrics of the heart function would, therefore, be extremely useful.
We will process cardiovascular MRI and patient data to assist in the formulation of a medical diagnosis. Using deep learning, image segmentation is performed on the MRI data such that metrics can be derived that potentially are predictive indicators for certain diseases and physical conditions. Patient data consists of variables such as age, gender, earlier diagnosis, medicine usage, etc.
In the hackathon, we focus on feature extraction from the segmentation masks and combining those features with patient data. In preparation for the hackathon, we will deliver the segmentation masks (the training of deep models is computationally not feasible for a single day session) and propose several use cases.