AI-based Evaluation of cardiac real-time MRI with congenital heart disease

Abstract

Cardiac MRI scans are an important tool used by cardiologists. It can be utilized for diagnoses and assessment of vital parameters like the stroke volume or blood flow. The significant reduction in acquisition time that is achievable nowadays allows the recording of realtime 3D videos of the human heart. Such high frame rates yield very large amounts of data. To evaluate this data highly efficiently and with as little manual intervention as possible, it must be processed and interpreted automatically. We introduce the key challenges in such automated evaluations of realtime cardiac MRI. As patients are able breath freely during the examination, the effect of respiration on parameters such as blood flow and stroke volume can be investigated. However, this leads to the challenge of respiratory synchronization. Also, as the 3D representation of the heart is acquired slice by slice, the cardiac cycles in those slices must be synchronized. Another challenge is the physiological segmentation of the images. Especially for pathological hearts (e.g. univentricular hearts) this is a great challenge as high accuracy and automation are demanded. Furthermore, we present our workflow to tackle those challenges and show and discuss first results. One result is a breath- and cardiac cycle synchronized segmented univentricular heart in 4D (3D+time).