During cardiac interventions, in order to facilitate a correct device deployment, the relationship between the anatomy and the position of the instruments has to be known and presented to the surgeon. Therefore, tools that can automatically identify the position and orientation of the ultrasound probe and catheters in X-ray images as well as automatic localization of anatomic structures of interest in these images are highly desirable. Today most of these tasks are solved manually or semi-automatically, however automating them would improve the accuracy and also reduce the amount of time the interventionalists spend on aligning the images.
Possible topics:
-GAN's for multi modal image synthesis
- AI for aortic valve detection in mitral trans esophageal acquisitions of the heart. The ultimate goal is automatic detection of anatomical structures running very fast (real-time if possible), so that the 3D data could automatic be presented in a way that makes sense to the physician, by rotating, cropping the 3D data based on the information detected in the images.
-3D segmentation of valves, full 3D segmentation of the mitral valve from echocardiographic data. U-Net or more advanced architecture to be considered
-3D tracking of structures of interest in echo recordings, visual transformer-based architectures to be investigated, temporal consistency to be enforced
- AI for catheter detection in the ultrasound volume and highlight of it in the image. The goal is to automatically detect catheter, delivery device and artificial devices (like MitraClip or artificial valves). Those objects give somewhat different echo signals but not always very clear nor complete. The task will be to automatically detect such objects within 3D ultrasound datasets.
- AI for ultrasound probe detection in X-ray images, the task will be to identify the location of the ultrasound probe but also its pose
More info: https://folk.ntnu.no/kiss/projects2022/Research_projects.pptx
Supervisors: Frank Lindseth, Gabriel Kiss (COMP/IDI)
[ Skjul beskrivelse ]
