Deep segmentation leverages geometric pose estimation in computer-aided total knee arthroplasty

Knee arthritis is a common joint disease that usually requires a total knee arthroplasty. There are multiple surgical variables that have a direct impact on the correct positioning of the implants, and an optimal combination of all these variables is the most challenging aspect of the procedure. Usually, preoperative planning using a computed tomography scan or magnetic resonance imaging helps the surgeon in deciding the most suitable resections to be made. This work is a proof of concept for a navigation system that supports the surgeon in following a preoperative plan. Existing solutions require costly sensors and special markers, fixed to the bones using additional incisions, which can interfere with the normal surgical flow. In contrast, the authors propose a computer-aided system that uses consumer RGB and depth cameras and do not require additional markers or tools to be tracked. They combine a deep learning approach for segmenting the bone surface with a recent registration algorithm for computing the pose of the navigation sensor with respect to the preoperative 3D model. Experimental validation using ex-vivo data shows that the method enables contactless pose estimation of the navigation sensor with the preoperative model, providing valuable information for guiding the surgeon during the medical procedure.

Automatic Camera Calibration Using Multiple Sets of Pairwise Correspondences

We propose a new method to add an uncalibrated node into a network of calibrated cameras using only pairwise point correspondences. While previous methods perform this task using triple correspondences, these are often difficult to establish when there is limited overlap between different views. In such challenging cases we must rely on pairwise correspondences and our solution becomes more advantageous. Our method includes an 11-point minimal solution for the intrinsic and extrinsic calibration of a camera from pairwise correspondences with other two calibrated cameras, and a new inlier selection framework that extends the traditional RANSAC family of algorithms to sampling across multiple datasets. Our method is validated on different application scenarios where a lack of triple correspondences might occur: addition of a new node to a camera network; calibration and motion estimation of a moving camera inside a camera network; and addition of views with limited overlap to a Structure-from-Motion model.