Computer Vision News - February 2020

3 Summary with Leo Joskowicz 11 "That’s a problem we’re actively working on and I see it as a much-needed future goal.” “Typical ly, to get a detection or classi f ication appl ication to work you need hundreds of thousands of images. For segmentation, it ’s several thousand. This is very expensive. A large company, such as the ones that have publ ished lately – Google, NVIDIA, Zebra Medical – can afford to col lect and pay cl inicians to do that, but we at universities don’t have that luxury, so we have to be smarter. As is wel l known, necessity is the mother of al l inventions. We have to make do with very few annotated datasets and overcome this using the model - based methods.” Leo recognises this as one of the biggest chal lenges nowadays. How do you automate the process with the required precision and accuracy, but with the fewest possible number of annotated scans? And i f it ’s not possible, can you at least let the cl inician know when there is a lack of conf idence in a decision? Looking ahead, Leo hopes we wi l l see the development of a methodology that wi l l quickly turn around high-qual ity solutions to address speci f ic cl inical problems. “You have to understand that medicine has many sub-specialties. There are several dozen organs and hundreds or even thousands of pathologies. There is no generic solution to address al l of them. You end up developing things for speci f ic pathologies and organs and the space of possible problems and solutions is very, very large.” Final ly, he points out that although attention is currently on detection and identi f ication solutions, there is another area of opportunity. “30 to 40 per cent of the images that a radiologist reads are not original images, but images that are taken after treatment was given,” he explains. “This is the fol low-up case. Now that you know that there is this pathology, what happens as things change over time? That ’s a problem we’re actively working on and I see it as a much-needed future goal .”