Computer Vision News - May 2016

Application COMPUTER VISION NEWS The key challenges to the creation of a technology which would answer the question were twofold: First, the physics of the problem at hand : measuring blood quantity within sponges and suction canisters. An easy, rapid and accurate measurement would prove itself very useful and, of all the approaches, computer vision is the one which is the best equipped to analyse the visual information embedded in the sponges, having a much better visual acuity than human observation, especially under varying ambient conditions, like lighting and room setting. Given the heterogeneous nature of the blood distribution on the sponges and the need to take advantage of the visual information available, this is a perfect problem for the computer vision approach. Second, the need for a great delivery system to use in the surgery room : a real-time vital sign measurement resulting of raw imaging data embedded in the sponges to be delivered in the era of mobile devices, with beautiful interfaces built with the user in mind. What Gauss Surgical did was to build an app that allows the user to scan the sponges in a simple manner, so that they can capture that data-rich image, upon which employ computer vision techniques to rapidly and precisely calculate blood quantity. The more soaked and red the sponges are, the more blood they contain. The first task of the software is therefore to obtain and use geometric features, like the size of the sponges and their mapping in real world coordinates; it then needs to look at the percentage of blood coverage of each sponge, applying segmentation and background removal techniques. Pixel- level information is standardized, taking into account variations like ambient lighting, so that geometric data narrows the photographic information. Following this data-heavy approach, the developers used all these features to build a large dataset of labelled images with which map the various features on known quantities of haemoglobin. 5