Computer Vision News - October 2020

RSIP Vision's Medical Projects 14 Tracking tissue movement - This task is one of the most challenging tasks in the field, as the deformation of tissue can take place in large areas and is very rapid. In many cases, parts of the tissue are occluded by different tools in the scene during the procedure . The solution in this case needs to be tailored to the specific needs, and in some cases classical approaches can work, while in other cases neural networks need to be used. One of the main challenges in these algorithms is the lack of ground truth. An additional challenge is ground truth annotation, and the learning in this case will be unsupervised learning. Algorithm used for the coordination between robot and camera Calibration - The calibration of the imaging device used in the procedure is a crucial factor in keeping the procedure precise. Calibration is a process that is unique for each system setup. For example, a system that contains a fluoro system (C-arm) will need to address both its internal and external calibration parameters, and in some cases (depending on the hardware) should also address distortion issues. Calibration is often performed by using a unique jig, using the data from the hardware, or by installing additional hardware on the camera. The decision of what type of calibration to use, and how to implement it, is based on the procedure precision requirements, the current hardware, and the method by which the procedure is performed. Registration - In many systems, several coordinate systems are used, and registration between these systems is required. For example, a pre-scan 3D image, such as CT or MRI, is used as a 3D Figure 4: Detection of cancer tissue during colonoscopy.