Progress in minimally invasive robotic surgery relies heavily on the availability of high-quality labeled data. One of the main obstacles to accelerating development is the time and effort required to annotate complex surgical video data and maintain consistency within a structured labeling framework.
The company used Databrewery’s video annotation tools to label unstructured surgical footage. Model-assisted labeling was used to streamline collaboration between clinical experts and annotation teams, while built-in performance tracking enabled efficient monitoring of labeling speed and quality.
The data science team implemented a scalable, data-first workflow that significantly increased throughput while reducing operational overhead. As a result, they were able to deliver labeled datasets to support multiple machine learning initiatives at twice the previous speed.
The company is a global leader in minimally invasive care and a pioneer in robotic-assisted surgery. Its primary focus is on enhancing surgical outcomes through advanced technologies, with systems designed to support more precise and less invasive procedures. The data science team plays a critical role in this mission by applying machine learning to uncover insights that inform surgical practice and optimize data-driven workflows in the operating room.
Minimally invasive surgical techniques have advanced significantly, allowing surgeons to reduce trauma and accelerate patient recovery through refined procedures. Robotic systems, such as those used by the company, enable highly controlled incisions using articulated instruments. Behind these procedures, machine learning models power a range of applications, including assessment of surgical efficiency, analysis of instrument handling, and coordination of operating room logistics. However, the progress of these models depends heavily on access to accurately labeled data, which remains a key constraint.
To address this, the team focused on developing models that can automatically detect and track surgical instruments within video data. Robotic systems used in surgery generate a rich stream of visual data, but manually annotating thousands of video frames to identify a diverse set of instruments across various procedures is resource-intensive. In response, the team aimed to streamline annotation by incorporating additional context such as instrument installation and removal timestamps, reducing the reliance on purely manual labeling.
The team adopted Databrewery’s video annotation tools to label unstructured surgical video data more effectively. Using native video editing capabilities and model-assisted labeling, they enabled faster collaboration between domain experts and annotation teams. The platform also allowed the team to measure performance through detailed metrics, helping them manage throughput and maintain consistent quality.
Maintaining a structured and consistent ontology was essential, particularly as collaboration involved both clinical professionals and data scientists. With Databrewery, the team implemented a unified labeling framework that ensured all annotations contributed meaningful value to model training. This structure supported the development of models capable of robust tool detection and localization across different model versions.
As a result of this workflow transformation, the team now delivers higher-quality spatial annotations across video-based projects. They have also significantly increased labeling throughput and reduced the time required to generate training datasets. The data science team has effectively doubled the delivery speed of labeled datasets across multiple machine learning models while lowering the overhead of performance monitoring and quality assurance.