Page 144 - Artificial Intelligence for Computational Modeling of the Heart

116 Chapter 3 Learning cardiac anatomy

speed. Using robust statistical shape modeling within the multi-
scale image navigation paradigm, a system can be defined, that
can accurately recognize whether any anatomical structures of in-
terest are missing from the field of view.
For the task of image segmentation, we show the performance
of the marginal space (deep) learning frameworks and of modern
image-to-image deep learning models in segmenting anatomi-
cal structures like the heart chambers or valves. Here we discuss
the advantage of image-to-image solutions in implicitly capturing
prior shape information and use it for a more robust prediction.
In the space of image tracking, we highlight in this chapter
how modern approaches in artificial intelligence have demon-
strated their abilities in encoding complex visual object features
and decoding their motion patterns both efficiently and effec-
tively. They have shown great potential towards real-time model-
ing for anatomy motion and deformation, and thus offer support
for accurate cardiac function assessment, anomaly detection and
prediction.