Deep Learning

As a partner in the BIA project INtelligent Cardiovascular Ultrasound Scanner - INCUS  we perform research on deep learning methods to increase productivity and accelerate decision making for ultrasound operators. Cardiac related disease is the number one cause of death in the Western world, including Norway. Echocardiography is the most important imaging tool for the cardiologist to assess cardiac function. An echo examination of the heart is real time, cost effective and can be performed without discomfort to the patient and without harmful radiation. These are great advantages compared to other medical imaging modalities.

Deep learning has shown great progress for 2D data especially for image classification and object detection. For 3D data, it is still early days, even though the amount of 3D data is in rapid growth. Our research aims at developing methods that are able to understand the content of both scanned and designed data, in particular for building information.

Google maps is an impressive example of what can be achieved with data driven prediction. Instead of carefully modeling each road and scenary, their methods rely on huge amounts of user data and carefully selected algorithms. The combination of recent advances in the field of deep learning and the widespread of sensors today, opens up the possibilty to make more accurate predictions than ever before, without the need of modeling the entire system.