Theator Uses NVIDIA Clara AGX to Develop Surgical Analytics
In recent years the healthcare industry has applied data analytics to offer higher quality care to patients, but one area that hasn’t yet caught up is the operating room (OR).
Hospitals and surgeons need better training models, access to experts, and data-based decision-making to level the field and decrease surgical outcome variability — and data analytics and artificial intelligence can help. Hospitals and healthcare systems already sit on tremendous troves of data that, if utilized correctly, can enable more accurate decision-making in the OR and during pre-and post-op treatments. The potential is there, but a major issue still remains — what happens in the OR basically stays in the OR. Intraoperative recordings are not yet mainstream, though we seem to be heading in that direction. But even if you have access to surgical video, it’s a formidable challenge to analyze each video frame by frame to annotate and render them useful for surgeons. The computing power required goes beyond the standard existing capabilities which today’s healthcare organizations possess and requires specialized software expertise for its successful implementation.
At Theator, we are harnessing the latest trends and technologies in the AI and computer vision space to fill these critical gaps. We look forward to using the NVIDIA Clara AGX developer kit to support our platform as we grow and develop.
Advanced computer vision techniques enable us to implement precise surgical phase recognition, which can discern relatively small differences throughout operational footage and significantly enhance the quality of surgical reviews. A hybrid AI system processes footage using a hybrid AI system that runs on GPU-powered on-premises edge devices and then uploads it to Theator’s cloud-based library, one of the world’s largest collections of annotated surgical video data, containing more than 10,000 hours of curated surgical video. We apply algorithmic capabilities to create a holistic Surgical Intelligence offering which not only captures surgical steps, events, safety milestones, and complexity levels but also analyzes correlations between operations.
Our technology can index large-scale surgical datasets with an accuracy rate vastly exceeding that of similar platforms. This creates a practical function for the near-limitless amounts of surgical video footage, which are recorded in the operating room but often remain largely siloed as individual surgeons or departments cannot be expected to analyze and organize this material at scale. Our platform collects insights and conducts edge analytics at scale to tag a large volume of surgeries frame by frame, a process enhanced by Theator’s Video Transformer Network (VTN), a pioneer transformer-based framework in video action recognition. The VTN generates significantly reduced training runtime (16.1X) and an accelerated inference process (5.1X) while maintaining accuracy levels when compared to competitors.
The advancements that NVIDIA Clara AGX offers the medical space are both impressive and necessary. Clara AGX increases efficiency in products and tools being used by people worldwide, remotely or otherwise. Real-time AI and advanced images, videos, and signal processing are essential to improving patient care and therefore saving lives. Securely orchestrating AI application deployments to fleets of medical devices or edge nodes will allow healthcare providers to rely on their equipment more easily and focus more heavily on developing treatment plans or diagnoses, also made more efficient by the implementation of AI.
To learn more about AI startups revolutionizing healthcare, join NVIDIA Inception and me at the NVIDIA GPU Technology Conference on Monday, April 12th, at 10:00 a.m. PDT.
- Dotan Asselmann, CTO and Co-Founder, Theator
