Computer Vision News - October 2020

AI for Surgical Robotics 13 algorithms is that they need to run in real time, and the system will not accept any delays caused by long computations. RSIP Vision offers a strong suite of software modules and capabilities for surgical applications. Below is a brief description of some of the algorithms that we have developed in this field. Our solutions have been integrated into industry leading products in this market. Visual Enhancement Image Segmentation - One of the basic tasks of scene understanding is to segment an image into different objects and components. This is done both in 2D images as well as 3D images. In medical applications, the fine details of the segmentation have major significance. The object that is segmented can be an anatomical part or tools that are seen in the image. In many cases, a U-net architecture is used as the starting point for the neural network, followed by a variety of steps that will improve the outcome of that module. We had successfully delivered segmentation modules to several customers that are using it as part of their procedures. Classification & detection - In many procedures there is a need to detect an object in the image, and in other caseswe will need to find the category of an object in the image. For example, detecting tools seen by the camera, followed by classifying the type of the tool seen in the camera, identifying the organ seen in the camera, or classifying the state of an object in the scene. In many cases there is more than a single instance of the object, and the classification is done after an instance segmentation. For state-of-the-art performance, neural networks areused to learn these complex classification tasks. Figure 1: Chest CT segmentation including airways, fissures, and lung lobes Figure 2: Shoulder joint bone segmentation Figure 3: 2d segmentation of knee joint in X-ray images