Computer Vision News - June 2020

3 Summary WEISS at UCL (part 2) 27 for pathology is at a very impressive level, ” he tells us. “Every time I hear a talk from Paige AI I’m just amazed at the sheer amount of data that they’re able to process together. With surgical data sets, we are more limited, so it’s much more challenging to have reliable algorithms.” Translating work from the research stage to the operating room can be a long process which has to be taken step by step. Work on proof of concept is tested in the lab before it moves to the stage of applying for a clinical trial. Francisco has experience here from his work on developing an ultrasound simulator for the purpose of training clinicians. “We had to do a study first with a phantom, then before the lockdown we were acquiring data from real patients so we could do retrospective analysis. Now we are applying for funding to have a standalone version of the simulator and then we will have to apply for more funding for clinical trials.” It’s not a device that will be used in actual surgery, so in theory the process should be more straightforward, but Francisco says there is still a way to go. “We started this three years ago and it’s still going. Even a simulation project for training, which is much simpler than a device for surgery, is a very long process.”