MICCAI 2019 Tuesday

MICCAI 2019 DAILY 9 Pablo Márquez Neila examples of potential applications, in problems like biomarker detection in OCT scans, cancer detection, cancer diagnosis, and segmentation of damaged tissue. Outside of the medical field, there are self-driving cars with autonomous driving systems that are trained to detect pedestrians. Sometimes there is a situation where the camera breaks, you have very adverse weather, or the cameras are receiving images that they weren’t trained for, and the system is not ready to detect those cases. Raphael tells us that one of the areas they thought initially could be very interesting for this work is the process of certifying technologies towards medical processes , where guarantees are required by notified bodies to ensure that technology is safe for the public. Any system that is trying to use deep learning technology to process their respective data would hypothetically benefit from the system checking that that data should actually be used on a real patient. He says that is where they see this kind of effort go forward. Pablo thinks they are opening a new field in this area, which seems to be unexplored by the community. Raphael adds that historically, we have looked at image quality as being the fundamental rule or metric to say something similar, but they are going far broader in saying that, regardless of your device, regardless of the technology of the sensors you’re using, this concept of validating the data is key. Pablo looks ahead to next steps: “We want to keep moving forward and keep trying to reach validation systems that can reach industrial-level performance . So far, we have a pretty good performance, but for a medical device in a hospital, for example, we need to push the performance higher. From a theoretical point of view, we would like to understand more about how this works and what the possible extensions of this idea are. For example, we have been thinking that this same idea could help to annotate datasets that are not annotated, to cover better the training space.” If you would like to find out more about Pablo and Raphael’s work, come along to their poster session [T-5-E-468] today at 13:00. "Outside of the medical field, there are self-driving cars with autonomous driving systems that are trained to detect pedestrians."