Bay Vision - Spring 2018

Andrew Hosford Siddarth Satish Kevin Miller Gauss Surgical is the only FDA-cleared iOS computer vision medical device for the operating room. Two years ago, Computer Vision News spoke to CEO and founder Siddarth Satish about their Triton mobile platform for real-time monitoring of surgical blood loss. Now VP of Engineering Drew Hosford , and Senior Algorithm Scientist Kevin Miller are bringing us up to speed with how far their technology has come. Giving doctors an accurate measure of blood loss allows better transfusion decisions. This can save lives, because in addition to reducing the number of transfusions, it can recognize hemorrhages during surgery. Across the world, especially in the US, there are an increasing number of hemorrhages during childbirth – more mums are dying in the US today from profuse blood loss than they were in the 1980s. Part of the reason for that is that doctors cannot easily recognize how much blood has been lost so it is hard to tell the difference between a hemorrhage and just normal blood loss. In the past, they would go off their gut feeling and if it looked like more blood than normal, they would transfuse. Drew says that using their solution, hemorrhage is recognized two to three times faster. Not only is it recognized sooner, but doctors are able to prevent hemorrhages from happening because they can see a trend and put in the proper protocols to ensure that the patient is taken care of before it gets worse. Their solution works by using photos of sponges and canisters taken by an iPad. Those sponges have a mixture of blood and saline. If it was pure blood, they could be weighed to give an accurate result, but as they are a mixture, Gauss use computer vision techniques to accurately determine the volume of blood loss . Proving they could do that was a bloody process indeed! They went to the blood bank and got expired banks of blood, threw in different combinations of blood and saline and imaged it in various lighting conditions, rotations, positions, and backgrounds to train their algorithm. Drew explains: “ We literally do all sorts of combinations of hemoglobin concentrations, combinations of blood with plasma, then pour it on the sponge, image it, pour more saline on it, mix it all up, image it again, pour more saline on it, mix it up, image it again. You do each image in multiple lighting conditions across multiple different backgrounds and positions in the operating room. That is one of the things that Kevin uses to train the model file and then of course we also have to validate it using that exact same process. It is actually a very bloody, time-intensive, labor-intensive process both in terms of training as well as in terms of validation. ” Kevin tell us more about the algorithm: “ We have an algorithm that takes a picture of a sponge, canister, or any bloodied substrate, and we estimate the amount of blood on the substrate so we can give an accurate estimate of the amount of blood that has been lost through the case. In terms of how the algorithm works, right now we are using traditional computer vision methods . Bay Vision 4 Boston Vision Gauss Surgical