41 Computer Vision News realistic rendering and speckles and changes in appearance, which are based specifically on the physical characteristics of the materials. Under the supervision of a surgical resident, we used Blender to design these scattering functions visually to match the appearance of those organs during surgery by using some open-source images. With such a large quantity of data and the availability of open source 2D video of surgery, it would have been nice to learn in a data-driven way what those scattering functions look like, so you can get as close to data realism as possible.” SARAMIS is a collaborative effort between WEISS and CMIC teams involving clinicians, labelers, and algorithm developers. Nina is based primarily in WEISS, but the team comprises different subgroups, including a clinician subgroup of practicing radiologists from UCLH hospitals and a labeling subgroup of mainly PhD students in medical imaging, with a lot of expertise in what anatomy looks like. The latter did the first pass of the labeling before sending the data to clinicians. The algorithms and dataset were primarily developed at WEISS, directed and organized by Nina. “I’d like to warmly invite the wider computer vision community to look at SARAMIS,” she says. “I think it’s quite interesting in terms of opening new avenues for improving vision algorithms in surgery and potentially opening the door to pure computer vision practitioners to the surgical setting. We’d really appreciate the feedback. Just drop me a line if you want to chat about it.” Beyond SARAMIS, Nina, originally from Spain and now based in London, is pursuing her PhD focused on developing automatic registration methods for trackerless image guidance in laparoscopic liver surgery. She is also an ML engineer at Tortus, developing an AI co-pilot for doctors, which helps with automated documentation and computer control of EHR systems. “I started my PhD just before the pandemic,” she reveals. “I was going to be in surgery collecting lots of data and doing supervised learning to enable these algorithms and methods. However, the pandemic happened, which threw a bit of a spanner in the works! We couldn’t go into surgery because all the hospitals got turned into Covid wards, so the data collection aspect was thrown out of the window. Instead, I pivoted my work to look at synthetic data generation, learning regimes that use synthetic data, and the potential of using algorithms in a synth-to-real approach.” SARAMIS
RkJQdWJsaXNoZXIy NTc3NzU=