Computer Vision News - December 2018

12 Computer Vision News Left Atria Reconstruction with Deep Learning Every month, Computer Vision News reviews a successful project. Our main purpose is to show how diverse image processing techniques contribute to solving technical challenges and real world constraints. This month we review RSIP Vision’s 3D Reconstruction and Deep Learning work in the field of cardiology: Left Atria Reconstruction from a Series of Sparse Catheter Paths Using Parametric Model and Neural Networks . This research is the result of a cooperation between RSIP Vision, a major industrial concern and the University of Tel Aviv . Project Modeling and reconstructing the shape of a heart chamber from partial or noisy data is useful in many minimally invasive heart procedures. Current solutions involve navigating a catheter to the heart chamber, where a map of the chamber itself is needed, with its whole three-dimensional shape. A magnetic field with special properties is created in the area and the catheter is equipped with number of electrodes measuring the magnetic fields, by which it is possible to know exactly where the electrode is and as a result where the catheter is or moves. Then you can acquire a point cloud and eventually construct a 3D map of the chamber. But this process is time consuming and also prone to errors, as the catheter itself can deform the walls of the chamber, which will make it difficult to understand the real shape. Solutions marketed until now obtain only a noisy map which is very difficult to understand and use. Here is where RSIP Vision’s expertise is needed: our algorithm is able to create a map of the heart chamber using sparse data. Even when we have only a sparse set of points, we can map the whole chamber. And even when the data is noisy, we use mathematical methods to reconstruct the shape of the chamber. As a consequence, we do not really need each and every point in the chamber: we can acquire a small set of key points and then construct the most probable shape, given this subset of points. The way we did it is by creating a parametric model which describes the shape using only a small set of properties. In this way, the shape is calculated and displayed in real time to the physician, including information about the position of the pulmonary veins and how curved they are. The Here is where RSIP Vision’s expertise is needed: our algorithm creates a map of the heart chamber using sparse data Side viewof reconstruction - Red: input volume in ground-truthwire frame; Green: reconstruction result.