Data-tailored shape compression for interventional device tracking

Fiber Optic RealShape (FORS) is a novel technology using embedded optical fiber, aiming to visualize the full 3D shape of interventional devices, such as catheters and guidewires in real time. Visualized within the context of anatomical data, often imaged by a combination of X-ray and contrast-enhanced CT (computed tomography), the 3D shape of the interventional device provides intuitive guidance to clinicians without being exposed to radiation and assists navigation of devices to challenging targets. The ability to store shape data during procedures is essential to gain technical insights and to improve the accuracy and robustness of the technology, which can help to improve certain operation aspects during surgery. However, continuous shape sensing can incur large data streams that can limit the amount of data stored during interventions and hamper the effectiveness of postoperative data analysis. This paper aims to alleviate the data demands of shape sensing devices by proposing a data compression algorithm. We investigate two cases: 1) compression when full samples are sensed at the encoding side, and 2) combined compression and optimization of sampling points for sensing. The presented study achieves these objectives through 1) using Principal Components Analysis (PCA), and 2) compressed sensing employing a data-tailored transform based on PCA. The developed intelligent compression algorithm can compress the position-based shape sensing data by retaining only 2% of the original data points while preserving high reconstruction quality of the shapes. When full samples are available, the proposed technique achieves a compression factor of 67.5 at acceptable loss levels.