MRI can feel like driving a high-tech car. As a driving assistant for navigation and parking we are developing a tracking and decision support system.

The patterns of markers are localized by unique deep learning in the MRI or CT image. The pattern of markers is immediately recognized by an algorithm trained on over 10,000 MRI images and validated in clinical interventional settings.

Our approach is generic and can be applied to any endovascular system (arterial, venous) along the aortic pathway and will support procedures from iliac angioplasty, vena cava filter removal up to supra-renal aneurysm repair.

Preclinical studies were performed in phantom and pig models. Overall performance of 88.6% detection rate with 1.1% false positives in the animal datasets.

Development with state-of-the-art techniques (cycle GAN) and a 3D point-cloud based translation of all possible markers orientations in real-life MRI images. Outcome: a decision support system to sense and steer the device.

Overall performance of 88.6% detection rate with 1.1% false positives in animal testing

(image courtesy UTSW Dallas and Fraunhofer MEVIS)