Computer Vision News - December 2022
49 PI-CAI: Prostate Cancer AI and to the best of our abilities, they include the entire spectrum of radiologists. Now, we have started to see that AI can make a big difference . ” Joeran adds: “ For me, the goal is to have an AI detection system that performs at least as well as expert-level radiologists! ” The team will not give too much away but tell us that preliminary results are promising. Another goal is to unify how this task is performed so that everyone agrees on what they want fromAI, how to evaluate it, what the datasets are, and how clinical and technical experts work together. The team also hopes it inspires others in the community to contribute to PI-CAI in the future, including proposing new datasets and tasks. Although it is too late to give anyone advice for submissions this year as the challenge is already in the testing phase, what do they hope participants have taken into account? value of AI, ” Jasper tells us. “ In a concurrent reading setup, we’ve seen that AI improves the workflow period, so how long it takes to do your reporting. It can also assist in annotating suspicious areas and providing suspicion scores. We’ve also seen in these studies that agreement among radiologists improves, with less experienced readers getting better and finding they are more likely to agree with the experts. ” Does the team expect the winning model to perform better than doctors? Anindo reveals that at the start, he was skeptical, based on prior studies that they have compared against expert radiologists, but his view has shifted. “ We have an estimate of how many cases are needed to train such a model and how well experts rank, ” he says. “ We’re doing a Reader Study in conjunction with PI-CAI, and it’s not just a handful of experts. We have 79 radiologists enlisted worldwide , From left to right: co-organizers Anindo Saha, Henkjan Huisman, Maarten de Rooij, Jasper Twilt, and Joeran Bosma this week at RSNA.
Made with FlippingBook