Surgery is one of the domains in which we’re seeing some of the most exciting AI innovations today. At the forefront of this work is Kaliber, a presenter at this year’s GTC and a member of NVIDIA Inception, an acceleration platform for AI, data science and HPC startups that provides go-to-market support, expertise, and technology. Kaliber is building AI skills and software solutions to assist and support surgeons performing arthroscopic and minimally invasive procedures. With its AI-powered surgical software suite, Kaliber is working to dramatically reduce surgical variability and, in doing so, improve patient outcomes, increase patient access, and redefine standards of care.
There is a need within the surgical community for AI-powered tools.
The goal of any surgeon is to consistently achieve optimal outcomes for patients. Patients want the same thing: superior care no matter where they live or who their surgeon is. However, surgical variability remains a significant problem–one that translates to surgical error, high healthcare costs, inequitable access to proficient care, and low patient confidence.
Addressing surgical variability, though, is complicated as it results from a number of intersecting factors that include not only innate ability, but also the quality of a surgeon’s training, the volume of procedures a surgeon has performed, access to (and the quality of) ongoing education and training, cognitive load, and information available. The root problem, by and large, is that there exist very different surgical career paths; some offer extensive opportunities for specialized training and practice, close mentorship, and ongoing education, while others provide far fewer of these opportunities. The result is a significant gap in surgical proficiency, one that oftentimes disadvantages more rural and poorer communities.
AI-assisted surgery, however, has the potential to address many of these challenges. By providing checklists, procedural affirmation, and advice and guidance before, during, and after surgery, AI-powered solutions act as a surgical co-pilot: providing guidance when consulted and caution when a surgeon strays from best practices and data-informed wisdom. Reproducibility manifested.
Deploying AI in medical instruments presents its own challenges.
From collecting and annotating data, to selecting and training the right model, to deploying that model in embedded systems in medical instruments, the systems and processes required to successfully integrate AI into surgical practice can be daunting. Moreover, AI skills must constantly be retrained with new data and updated, validated, and re-deployed to ensure that surgeons always have ready access to a set of standardized and fully up-to-date AI skills.
Realizing the dream of AI-assisted surgeries requires solutions to several complex technical problems: a multi-functional cloud repository for storing updated AI skills; a mechanism that can move AI skills to the edge while maintaining security and safeguarding patient information (PII); a runtime for the edge device that allows it to perform skills with extremely low-latency; and a real-time intraoperative AI platform that interprets the ever changing surgical field of view.
Additionally, intraoperative solutions must be FDA cleared before surgeons can interact with them.
Kaliber is finding solutions to these challenges and using AI to build the tools that surgeons need now.
Kaliber is developing a first-of-its-kind AI-powered surgical toolkit that marries the skills of the surgeon with the precision and accuracy of AI-powered information and guidance systems. Kaliber’s solutions address pre-operative, intraoperative, and post-operative needs and include surgical planning and patient communication support, intraoperative guidance systems, real-time recognition of anatomy, pathology, tools, and activities, bluDot™ pinless landmarking, and digital measurement.
NVIDIA GPU-optimized software AI and the NVIDIA Clara AGX developer kit have been instrumental in Kaliber’s work. The technical foundation of all of Kaliber’s software solutions can be found in its surgical AI platform, medical data unifying system architecture (MeDUSA), which is built on top of NVIDIA’s AI hardware and software stacks. MeDUSA eliminates bottlenecks with an end-to-end, industry-grade, AI development workflow, which operates as follows:
High volume raw medical data ingestion
Catalog and organize unlabeled datasets, AI-assisted
AI-assisted labeling tools
Train and produce models from labeled datasets
Catalog and store model generations for traceability
Validate and test models for accuracy and speed
Convert models to NVIDIA TensorRT engines
Deploy validated NVIDIA TensorRT engines to edge devices
MeDUSA allows developers to scale the Medical AI workflow by providing prebuilt skills that act as building blocks for Kaliber’s suite of software solutions.
These solutions are hosted in the cloud and then deployed using NVIDIA Clara AGX, an embedded AI computer and software development framework designed specifically for medical devices that require the ability to perform real-time AI and advanced image, video, and signal processing. Clara AGX has been an invaluable tool in Kaliber’s work; it functions as a plug-and-play edge device and enables AI in legacy medical equipment via its onboard HDMI capture card. With Clara AGX, surgeons essentially have access to a catalog of Kaliber’s surgical skills, which they can call upon as needed for specific procedures.
Perhaps the most critical contribution of Clara AGX, however, stems from its ability to deliver real-time inferencing speeds with its Quadro RTX 6000 discrete GPU. For AI to be successful in intraoperative applications, it must be delivered in real time; with Clara AGX, Kaliber has been able to achieve real-time delivery in the OR with inferencing speeds as low as 1.25ms utilizing NVIDIA TensorRT and CUDA optimizations. Importantly, Clara AGX also offers ample computational headroom to accommodate multiple models running simultaneously (for example, digital measurement and bluDot™).
Kaliber and NVIDIA are redefining what’s possible in minimally invasive surgery.
Kaliber’s collaboration with NVIDIA has opened up new possibilities for the use of AI in surgery. While Kaliber is currently focused on building solutions specifically for arthroscopic surgery, it also plans to offer MeDUSA and KL Inference in the form of an SDK so that other companies can build custom offerings for different specialties. With its own suite of surgical software solutions as well as the forthcoming SDK, Kaliber is helping build an ecosystem of innovative surgical technology companies and equipping surgeons across specialties with the tools they need to reduce surgical variability and to improve patient outcomes and access to proficient care.
For more information, visit kaliber.ai and view our session at GTC “Platform to Deliver Enhanced Surgical Outcomes with AI-Powered Medical Devices” [S23054]. Registration for the event is free of charge.
About Kaliber Labs
Kaliber builds AI-powered solutions to meet the needs of patients, surgeons, and administrators in arthroscopic surgery. Our intelligent software is designed to generate insights from surgical procedures—cut to close—and channel them into surgeon-centric tools to support quality, access, and cost of care.